fix(ci): strip free-form text and sanitize metadata on skipped JUnit elements

perf(editor): avoid brush cursor style invalidation

.github/scripts/check_test_execution.py

@@ -0,0 +1,17 @@
+"""Reject skipped/empty native CI runs without conflating driver child counts."""
+import argparse
+from pathlib import Path
+import xml.etree.ElementTree as ET
+
+parser = argparse.ArgumentParser(description=__doc__)
+parser.add_argument('report', type=Path)
+parser.add_argument('--minimum', type=int, required=True)
+args = parser.parse_args()
+root = ET.parse(args.report).getroot()
+cases = list(root.iter('testcase'))
+if len(cases) < args.minimum:
+    raise SystemExit(f'Incomplete execution: {len(cases)} cases, expected at least {args.minimum}')
+for case in cases:
+    if any(case.find(tag) is not None for tag in ('skipped', 'failure', 'error')):
+        raise SystemExit(f'Case did not pass: {case.get("classname")}::{case.get("name")}')
+print(f'Verified {len(cases)} executed cases; no skipped, failed or errored cases')

.github/scripts/run_native_tests.mjs

@@ -0,0 +1,46 @@
+import fs from 'node:fs';
+import path from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { spawnSync } from 'node:child_process';
+import { parse } from 'acorn';
+
+const root = fileURLToPath(new URL('../../', import.meta.url));
+const manifest = JSON.parse(fs.readFileSync(path.join(root, 'tests/native-ci-manifest.json'), 'utf8'));
+function files(directory) {
+  return fs.readdirSync(directory, { withFileTypes: true }).flatMap(entry => {
+    const target = path.join(directory, entry.name);
+    return entry.isDirectory() ? files(target) : /\.(?:mjs|cjs|js)$/.test(entry.name) ? [target] : [];
+  });
+}
+function importsNativeTest(node) {
+  if (!node || typeof node !== 'object') return false;
+  if (node.type === 'ImportDeclaration' && node.source.value === 'node:test') return true;
+  if (node.type === 'CallExpression' && node.callee.name === 'require' && node.arguments[0]?.value === 'node:test') return true;
+  return Object.values(node).some(value => Array.isArray(value) ? value.some(importsNativeTest) : importsNativeTest(value));
+}
+const discovered = files(path.join(root, 'tests')).filter(file => {
+  const source = fs.readFileSync(file, 'utf8');
+  return source.includes('node:test') && importsNativeTest(parse(source, { ecmaVersion: 'latest', sourceType: 'module' }));
+}).map(file => path.relative(root, file).split(path.sep).join('/')).sort();
+const assigned = Object.values(manifest).flatMap(group => Array.isArray(group) ? group : Object.keys(group));
+if (new Set(assigned).size !== assigned.length || JSON.stringify([...assigned].sort()) !== JSON.stringify(discovered)) {
+  throw new Error('Native test coverage must be exhaustive and disjoint; update tests/native-ci-manifest.json for added, removed or duplicated files');
+}
+for (const wrappers of Object.values(manifest['python-ci'])) {
+  for (const wrapper of wrappers) if (!fs.existsSync(path.join(root, wrapper))) throw new Error(`Missing pytest wrapper: ${wrapper}`);
+}
+if (process.argv.includes('--list')) {
+  console.log(JSON.stringify({ discovered, runners: manifest }, null, 2));
+} else {
+  const reports = path.join(root, 'test-results/native-node');
+  fs.mkdirSync(reports, { recursive: true });
+  const result = spawnSync(process.execPath, ['--test', '--test-concurrency=1',
+    '--test-reporter=spec', '--test-reporter-destination=stdout',
+    '--test-reporter=junit', `--test-reporter-destination=${path.join(reports, 'junit.xml')}`,
+    ...manifest['native-ci']], {
+    cwd: root, stdio: 'inherit',
+    env: { ...process.env, ENDPOINT_URL: 'http://127.0.0.1:1' },
+  });
+  if (result.error) throw result.error;
+  process.exitCode = result.status ?? 1;
+}

.github/scripts/verify_sanitized_artifacts.py

@@ -0,0 +1,465 @@
+#!/usr/bin/env python3
+"""Fail-closed independent verification and sanitization gate for test diagnostics.
+
+Runs after test suite completion or abnormal runner termination (e.g. failures, timeouts, crashes).
+Ensures that no unsanitized traces, raw credentials, temporary archives, or unauthorized
+diagnostics can ever be published as CI artifacts, while preserving safe JUnit XML reporting.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import logging
+import os
+import re
+import shutil
+import sys
+import tempfile
+import xml.etree.ElementTree as ET
+from pathlib import Path
+
+# Add repository root to Python path to import sanitizer
+REPO_ROOT = Path(__file__).resolve().parents[2]
+if str(REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPO_ROOT))
+
+try:
+    from scripts.sanitize_trace_artifacts import sanitize_trace_archive
+except ImportError:
+    # If import fails, define fail-closed fallback
+    def sanitize_trace_archive(zp: str) -> bool:
+        p = Path(zp)
+        if p.exists():
+            try:
+                p.unlink(missing_ok=True)
+            except OSError:
+                pass
+        return False
+
+logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
+logger = logging.getLogger("verify_sanitized_artifacts")
+
+PNG_MAGIC = b"\x89PNG\r\n\x1a\n"
+
+ALLOWED_JUNIT_ROOT_TAGS = frozenset({"testsuites", "testsuite"})
+ALLOWED_PUBLISHED_JUNIT_TAGS = frozenset({"testsuites", "testsuite", "testcase", "failure", "error", "skipped"})
+DISALLOWED_JUNIT_TAGS = frozenset({"properties", "property", "system-out", "system-err"})
+
+ALLOWED_ATTRIBUTES_BY_TAG = {
+    "testsuites": frozenset({"name", "tests", "failures", "errors", "skipped", "time", "id", "disabled"}),
+    "testsuite": frozenset({
+        "name", "tests", "failures", "errors", "skipped", "time", "timestamp", "hostname",
+        "id", "package", "disabled", "file"
+    }),
+    "testcase": frozenset({"name", "classname", "time", "file", "line", "assertions", "status"}),
+    "failure": frozenset({"message", "type"}),
+    "error": frozenset({"message", "type"}),
+    "skipped": frozenset({"message", "type"}),
+}
+
+SAFE_SKIPPED_STATUSES = frozenset({"skipped", "todo", "ignored", "disabled", "pending", "omitted", "test skipped"})
+
+SENSITIVE_KEY_PATTERN = re.compile(
+    r"(?i)(password|passwd|secret|token|api_key|apikey|access_token|refresh_token|"
+    r"private_key|auth_token|client_secret|session|session_id|sid|csrf|xsrf|jwt|credential|authorization)"
+)
+
+SECRET_REDACTION_PATTERNS = [
+    # 1. Authorization headers (Bearer / Basic)
+    re.compile(r"(?i)\b(bearer|basic)\s+([^\s,;\'\"<>]+)"),
+    # 2. Sensitive key-value pairs (key=value or key: value)
+    re.compile(
+        r"(?i)\b([a-z0-9_.-]*(?:password|passwd|secret|token|api_key|apikey|access_token|refresh_token|"
+        r"auth_token|client_secret|session|session_id|sid|csrf|xsrf|jwt|credential|authorization)[a-z0-9_.-]*)"
+        r"\s*([:=])\s*([^\s,;\'\"<>]+)"
+    ),
+    # 3. URL credentials (scheme://user:pass@host)
+    re.compile(r"([a-zA-Z][a-zA-Z0-9+.-]*://[^:\s]+):([^@\s]+)@"),
+    # 4. JSON-formatted key-value pairs
+    re.compile(
+        r'(?i)("?(?:[a-z0-9_.-]*(?:password|passwd|secret|token|key|auth)[a-z0-9_.-]*)"?\s*:\s*")([^"]+)(")'
+    ),
+    # 5. Synthetic secret keywords or tokens
+    re.compile(r"(?i)\b[a-z0-9_.-]*synthetic[-_]?[a-z0-9_.-]*\b"),
+]
+
+
+def safe_remove(filepath: str | Path) -> bool:
+    """Safely remove a file or directory fail-closed.
+
+    Returns True if the target does not exist or was successfully removed.
+    Returns False if removal failed and the target still exists on disk.
+    """
+    p = Path(filepath)
+    if not p.exists():
+        return True
+    try:
+        if p.is_dir():
+            shutil.rmtree(p, ignore_errors=False)
+        else:
+            p.unlink(missing_ok=True)
+    except OSError as exc:
+        logger.error("Failed to delete unauthorized file %s: %s", filepath, exc)
+        return False
+
+    if p.exists():
+        logger.error("Target %s still exists on disk after deletion attempt", filepath)
+        return False
+    return True
+
+
+def sanitize_text_content(text: str) -> tuple[str, bool]:
+    """Scrub sensitive credentials, tokens, and synthetic secrets from text."""
+    if not text or not isinstance(text, str):
+        return text, False
+
+    modified = False
+    out = text
+
+    # 1. Replace Bearer / Basic tokens
+    def _sub_auth(m: re.Match) -> str:
+        nonlocal modified
+        modified = True
+        return f"{m.group(1)} [REDACTED]"
+
+    out = SECRET_REDACTION_PATTERNS[0].sub(_sub_auth, out)
+
+    # 2. Replace key-value pairs
+    def _sub_kv(m: re.Match) -> str:
+        nonlocal modified
+        modified = True
+        return f"{m.group(1)}{m.group(2)}[REDACTED]"
+
+    out = SECRET_REDACTION_PATTERNS[1].sub(_sub_kv, out)
+
+    # 3. Replace URL credentials
+    def _sub_url(m: re.Match) -> str:
+        nonlocal modified
+        modified = True
+        return f"{m.group(1)}:[REDACTED]@"
+
+    out = SECRET_REDACTION_PATTERNS[2].sub(_sub_url, out)
+
+    # 4. Replace JSON key-value secrets
+    def _sub_json(m: re.Match) -> str:
+        nonlocal modified
+        modified = True
+        return f"{m.group(1)}[REDACTED]{m.group(3)}"
+
+    out = SECRET_REDACTION_PATTERNS[3].sub(_sub_json, out)
+
+    # 5. Replace synthetic secret markers
+    def _sub_synthetic(m: re.Match) -> str:
+        nonlocal modified
+        modified = True
+        return "[REDACTED]"
+
+    out = SECRET_REDACTION_PATTERNS[4].sub(_sub_synthetic, out)
+
+    return out, modified
+
+
+def prune_and_sanitize_junit_element(elem: ET.Element) -> bool:
+    """Recursively enforce restrictive published-JUnit representation.
+
+    Removes properties, property entries, system-out, system-err, and any
+    unnecessary free-form content. Preserves test identities, counts, outcomes,
+    and safe failure classifications.
+    Returns True if any modifications were made.
+    """
+    modified = False
+
+    # 1. Prune child elements not permitted in restrictive published representation
+    for child in list(elem):
+        tag_lower = child.tag.lower()
+        if tag_lower in DISALLOWED_JUNIT_TAGS or tag_lower not in ALLOWED_PUBLISHED_JUNIT_TAGS:
+            elem.remove(child)
+            modified = True
+        else:
+            if prune_and_sanitize_junit_element(child):
+                modified = True
+
+    tag_lower = elem.tag.lower()
+
+    # 2. Prune attributes not permitted in restrictive published representation
+    allowed_attrs = ALLOWED_ATTRIBUTES_BY_TAG.get(tag_lower, frozenset())
+    for attr_name in list(elem.attrib.keys()):
+        attr_lower = attr_name.lower()
+        if attr_lower not in allowed_attrs:
+            del elem.attrib[attr_name]
+            modified = True
+        elif tag_lower == "skipped":
+            val = elem.attrib[attr_name].strip()
+            if val.lower() in SAFE_SKIPPED_STATUSES:
+                if elem.attrib[attr_name] != val.lower():
+                    elem.attrib[attr_name] = val.lower()
+                    modified = True
+            else:
+                elem.attrib[attr_name] = "skipped"
+                modified = True
+        else:
+            val = elem.attrib[attr_name]
+            if isinstance(val, str):
+                new_val, changed = sanitize_text_content(val)
+                if changed:
+                    elem.attrib[attr_name] = new_val
+                    modified = True
+
+    # 3. Handle text content: structural and skipped elements do not carry free-form text
+    if tag_lower in ("testsuites", "testsuite", "testcase", "skipped"):
+        if elem.text:
+            elem.text = None
+            modified = True
+    elif tag_lower in ("failure", "error"):
+        if elem.text:
+            new_text, changed = sanitize_text_content(elem.text)
+            if changed:
+                elem.text = new_text
+                modified = True
+
+    # 4. Clean up tails: strip any non-whitespace characters from tails to prevent leaks
+    if elem.tail and elem.tail.strip():
+        cleaned_tail = re.sub(r"\S+", "", elem.tail)
+        if cleaned_tail != elem.tail:
+            elem.tail = cleaned_tail
+            modified = True
+
+    return modified
+
+
+def sanitize_and_verify_junit_xml(filepath: str | Path) -> bool:
+    """Verify that an XML file is an authentic JUnit report and enforce published representation.
+
+    Returns True if the XML file is a valid recognized JUnit report and was safely sanitized.
+    Returns False if the file is invalid, unrecognized, corrupted, or failed sanitization.
+    """
+    p = Path(filepath)
+    name_lower = p.name.lower()
+
+    # 1. Filename restriction: must match standard JUnit report naming
+    if not (name_lower == "junit.xml" or name_lower == "browser-junit.xml" or name_lower.endswith("-junit.xml") or name_lower.endswith(".junit.xml")):
+        logger.warning("Unrecognized XML report filename %s; not an authorized JUnit report", filepath)
+        return False
+
+    # 2. Parse XML
+    try:
+        tree = ET.parse(p)
+    except Exception as exc:
+        logger.warning("Corrupted XML report %s: %s", filepath, exc)
+        return False
+
+    root = tree.getroot()
+    # 3. Root tag restriction
+    if root.tag.lower() not in ALLOWED_JUNIT_ROOT_TAGS:
+        logger.warning("XML report %s has invalid root element <%s>; not a JUnit report", filepath, root.tag)
+        return False
+
+    # 4. Enforce restrictive published representation on root and all descendants
+    prune_and_sanitize_junit_element(root)
+
+    # 5. Normalize and rewrite sanitized report
+    try:
+        tree.write(str(p), encoding="utf-8", xml_declaration=True)
+        # Verify rewritten XML parses cleanly
+        ET.parse(p)
+    except Exception as exc:
+        logger.error("Failed to write sanitized XML report %s: %s", filepath, exc)
+        return False
+
+    return True
+
+
+def is_valid_disposable_ci_fixture(target_dir: str | Path | None = None) -> bool:
+    """Verify that execution ran inside an authorized disposable synthetic fixture."""
+    is_ci = bool(os.environ.get("CI") or os.environ.get("GITHUB_ACTIONS"))
+    if not is_ci:
+        # Local runs require explicit trace opt-in
+        trace_opt_in = os.environ.get("PLAYWRIGHT_ENABLE_TRACE", "").lower() in ("1", "true")
+        return trace_opt_in
+
+    if os.environ.get("ODYSSEUS_E2E_DISPOSABLE_ARTIFACTS") != "1":
+        return False
+    if os.environ.get("PHOTO_EDITOR_E2E_AUTH") == "0":
+        return False
+
+    allowed_tmp = [os.path.abspath(p) for p in (tempfile.gettempdir(), "/tmp", "/var/tmp")]
+
+    # Check 1: Verify fixture ownership via PHOTO_EDITOR_E2E_DATA_DIR
+    data_dir = os.environ.get("PHOTO_EDITOR_E2E_DATA_DIR")
+    if data_dir:
+        resolved = os.path.abspath(data_dir)
+        base_name = os.path.basename(resolved)
+        is_tmp = any(resolved.startswith(b + os.sep) or resolved == b for b in allowed_tmp)
+        if base_name.startswith("odysseus-photo-editor-e2e-") and is_tmp:
+            expected_db = os.path.join(resolved, "app.db")
+            db_path = os.environ.get("PHOTO_EDITOR_E2E_DB_PATH")
+            if not db_path or db_path == expected_db:
+                marker_path = os.path.join(resolved, ".disposable-ci-fixture.json")
+                if os.path.exists(marker_path):
+                    try:
+                        with open(marker_path, "r", encoding="utf-8") as f:
+                            fixture = json.load(f)
+                            if (
+                                fixture.get("synthetic") is True
+                                and fixture.get("authenticated") is True
+                                and fixture.get("database") == expected_db
+                            ):
+                                return True
+                    except Exception:
+                        pass
+
+    # Check 2: Verify fixture proof in target_dir if provided
+    if target_dir:
+        target_marker = os.path.join(str(target_dir), ".disposable-ci-fixture.json")
+        if os.path.exists(target_marker):
+            try:
+                with open(target_marker, "r", encoding="utf-8") as f:
+                    fixture = json.load(f)
+                    db_in_fixture = fixture.get("database", "")
+                    resolved_db = os.path.abspath(db_in_fixture)
+                    db_dir = os.path.dirname(resolved_db)
+                    db_base = os.path.basename(db_dir)
+                    is_tmp = any(db_dir.startswith(b + os.sep) or db_dir == b for b in allowed_tmp)
+                    if (
+                        fixture.get("synthetic") is True
+                        and fixture.get("authenticated") is True
+                        and db_base.startswith("odysseus-photo-editor-e2e-")
+                        and is_tmp
+                        and os.path.basename(resolved_db) == "app.db"
+                    ):
+                        return True
+            except Exception:
+                pass
+
+    return False
+
+
+def verify_and_sanitize_artifacts(target_dir: str) -> bool:
+    """Independently verify and sanitize diagnostics fail-closed, preserving safe JUnit XML.
+
+    Returns True if verification succeeded and artifacts are safe for publication.
+    Returns False if verification failed or deletion could not be confirmed.
+    """
+    if not os.path.exists(target_dir):
+        logger.info("Target directory %s does not exist; nothing to sanitize.", target_dir)
+        return True
+
+    disposable_ci = is_valid_disposable_ci_fixture(target_dir)
+    preserved_xmls = 0
+    sanitized_archives = 0
+    purged_files = 0
+    overall_success = True
+
+    try:
+        for root, dirs, files in os.walk(target_dir, topdown=False):
+            for filename in files:
+                filepath = os.path.join(root, filename)
+                name_lower = filename.lower()
+
+                # Cleanup disposable fixture marker so it is not published
+                if filename == ".disposable-ci-fixture.json":
+                    if not safe_remove(filepath):
+                        overall_success = False
+                    continue
+
+                # 1. XML reports: verify JUnit structure and sanitize credentials
+                if name_lower.endswith(".xml"):
+                    if sanitize_and_verify_junit_xml(filepath):
+                        preserved_xmls += 1
+                        continue
+                    else:
+                        logger.warning("Rejecting invalid/unrecognized XML report %s; purging fail-closed", filepath)
+                        if not safe_remove(filepath):
+                            overall_success = False
+                        purged_files += 1
+                        continue
+
+                # 2. Check if diagnostics are authorized (disposable synthetic fixture)
+                if not disposable_ci:
+                    logger.info("Purging diagnostic artifact %s (disposable CI fixture not verified)", filepath)
+                    if not safe_remove(filepath):
+                        overall_success = False
+                    purged_files += 1
+                    continue
+
+                # 3. Disposable CI is authorized: enforce strict sanitization
+                if name_lower.endswith(".zip"):
+                    ok = sanitize_trace_archive(filepath)
+                    if ok and os.path.exists(filepath):
+                        sanitized_archives += 1
+                    else:
+                        logger.warning("Trace archive %s failed sanitization; purging fail-closed", filepath)
+                        if not safe_remove(filepath):
+                            overall_success = False
+                        purged_files += 1
+
+                elif name_lower.endswith(".png"):
+                    # Validate PNG magic bytes to ensure not a secret payload disguised as an image
+                    is_png = False
+                    try:
+                        with open(filepath, "rb") as pf:
+                            header = pf.read(8)
+                            if header == PNG_MAGIC:
+                                is_png = True
+                    except Exception:
+                        is_png = False
+
+                    if is_png:
+                        continue
+
+                    logger.warning("Invalid or suspicious screenshot %s; purging fail-closed", filepath)
+                    if not safe_remove(filepath):
+                        overall_success = False
+                    purged_files += 1
+
+                else:
+                    # Purge any .tmp, .har, .db, .sqlite, .log or unrecognized files
+                    logger.info("Purging unauthorized diagnostic file %s", filepath)
+                    if not safe_remove(filepath):
+                        overall_success = False
+                    purged_files += 1
+
+            # Remove empty subdirectories
+            for d in dirs:
+                dirpath = os.path.join(root, d)
+                try:
+                    if not os.listdir(dirpath):
+                        os.rmdir(dirpath)
+                except OSError:
+                    pass
+
+        if not overall_success:
+            logger.error("Verification failed: one or more unauthorized files could not be purged.")
+            return False
+
+        logger.info(
+            "Artifact verification complete for %s: preserved %d XML reports, sanitized %d archives, purged %d files.",
+            target_dir,
+            preserved_xmls,
+            sanitized_archives,
+            purged_files,
+        )
+        return True
+
+    except Exception as exc:
+        logger.error("Unexpected error during artifact verification: %s; purging all non-XML files fail-closed", exc)
+        try:
+            for p in Path(target_dir).rglob("*"):
+                if p.is_file() and not p.name.lower().endswith(".xml"):
+                    safe_remove(p)
+        except Exception:
+            pass
+        return False
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("target_dir", nargs="?", default="test-results", help="Directory containing test results")
+    args = parser.parse_args()
+
+    ok = verify_and_sanitize_artifacts(args.target_dir)
+    sys.exit(0 if ok else 1)
+
+
+if __name__ == "__main__":
+    main()

.github/workflows/ci.yml

@@ -256,3 +256,10 @@ jobs:
         env:
           PYTEST_SHARD: ${{ matrix.shard }}
         run: python -m pytest -q -rs --durations=25 --shard "$PYTEST_SHARD"
+
+      - name: Playwright mobile layout checks (shard 1/4 only)
+        if: steps.docs-check.outputs.docs_only != 'true' && matrix.shard == '1/4'
+        run: |
+          ODYSSEUS_TEST_PYTHON=python ODYSSEUS_LAYOUT_BROWSER=chromium npx --no-install playwright test \
+            --config tests/e2e/ios-layout.config.cjs \
+            --reporter=line

.github/workflows/native-test-suites.yml

@@ -0,0 +1,129 @@
+name: Native test suites
+
+on:
+  pull_request:
+  push:
+    branches: [main, dev]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: native-tests-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  native-node:
+    name: Native Node regressions
+    runs-on: ubuntu-24.04
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
+        with:
+          python-version: '3.11'
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
+        with:
+          node-version: '22.23.4'
+          cache: npm
+      - run: npm ci --ignore-scripts --no-audit --no-fund
+      - uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: ~/.cache/ms-playwright
+          key: ms-playwright-${{ runner.os }}-${{ hashFiles('package-lock.json') }}
+      - run: npx playwright install --with-deps chromium
+      - name: Validate exhaustive disjoint runner mapping and execute omitted suites
+        run: |
+          node .github/scripts/run_native_tests.mjs
+          python .github/scripts/check_test_execution.py test-results/native-node/junit.xml --minimum 43
+      - name: Verify and sanitize native test results fail-closed
+        id: verify-native-results
+        if: always()
+        run: |
+          python .github/scripts/verify_sanitized_artifacts.py test-results/native-node/
+      - name: Save native test identities and results
+        if: always() && steps.verify-native-results.outcome == 'success'
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: native-node-results
+          path: test-results/native-node/
+          retention-days: 7
+          if-no-files-found: warn
+
+  browser:
+    name: Chromium application regressions
+    runs-on: ubuntu-24.04
+    timeout-minutes: 30
+    env:
+      ODYSSEUS_E2E_DISPOSABLE_ARTIFACTS: '1'
+      ODYSSEUS_TEST_PYTHON: python
+      CHROMADB_HOST: 127.0.0.1
+      CHROMADB_PORT: '1'
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
+        with:
+          python-version: '3.11'
+      - name: Install FFmpeg and establish system trust
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends ffmpeg ca-certificates
+          python scripts/check_ca_trust.py --system
+      - uses: astral-sh/setup-uv@c18668ad3cf93ea998bef934396af7bb5c839dc7 # v10.2.0
+        with:
+          version: '0.13.5'
+          enable-cache: true
+          cache-dependency-glob: requirements.txt
+          prune-cache: false
+      - run: |
+          uv pip install --system --require-hashes --only-binary=:all: --no-deps -r requirements-bootstrap.txt
+          python scripts/check_ca_trust.py --python
+          uv pip install --system -r requirements.txt
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
+        with:
+          node-version: '22.23.4'
+          cache: npm
+      - run: npm ci --ignore-scripts --no-audit --no-fund
+      - uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: ~/.cache/ms-playwright
+          key: ms-playwright-${{ runner.os }}-${{ hashFiles('package-lock.json') }}
+      - run: npx playwright install --with-deps chromium
+      - name: Run complete Chromium suite against a fresh authenticated app
+        env:
+          PLAYWRIGHT_JUNIT_OUTPUT_FILE: test-results/browser-junit.xml
+        run: |
+          npm run test:photo-editor -- --reporter=line,junit
+          python .github/scripts/check_test_execution.py test-results/browser-junit.xml --minimum 137
+      - name: Verify and sanitize failure diagnostics fail-closed
+        id: verify-browser-results
+        if: always()
+        run: |
+          python .github/scripts/verify_sanitized_artifacts.py test-results/
+      - name: Save per-test results and sanitized failure diagnostics
+        if: always() && steps.verify-browser-results.outcome == 'success'
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: chromium-regression-results
+          path: test-results/
+          retention-days: 7
+          if-no-files-found: warn
+
+  authoritative-native-suites:
+    name: Authoritative native suites
+    if: always()
+    needs: [native-node, browser]
+    runs-on: ubuntu-24.04
+    steps:
+      - name: Require both complete suites to succeed
+        env:
+          NODE_RESULT: ${{ needs.native-node.result }}
+          BROWSER_RESULT: ${{ needs.browser.result }}
+        run: |
+          test "$NODE_RESULT" = success
+          test "$BROWSER_RESULT" = success

docs/maintainers/native-test-coverage.md

@@ -0,0 +1,46 @@
+# Native test merge gate
+
+`Native test suites` runs on pull requests, dev/main pushes and manual dispatch.
+`Authoritative native suites` fails unless both the native Node and complete
+Chromium jobs succeed; it has no continue-on-error, retries or path bypass.
+Repository administrators must require this check alongside the four pytest
+checks in branch protection. The audited public dev branch had no protection or
+ruleset, so merely adding workflow jobs cannot prevent a manual merge.
+
+`tests/native-ci-manifest.json` exhaustively partitions all native Node files:
+16 run in the native job,14 through existing pytest wrappers, one through the
+existing frontend architecture CI step, and the formerly live email-reply file
+through an explicitly identified Playwright driver. The manifest validator fails
+for omitted/duplicate/unreachable entries. Node emits individual JUnit identities
+and propagates its failure exit. No native file is newly run in each pytest shard.
+
+The browser job uses the existing one-worker Chromium config and all 136 Photo
+Editor cases (including cursor and brush regressions), plus one driver for the native
+email-reply contract (137 cases total). Its ephemeral account,
+SQLite DB and server are created by the config/setup; reuseExistingServer remains
+false. NPX startup is offline/cache-only in a private cache; Notes/editor behavior
+and Python MCP services remain real. No developer app or external account is used.
+
+Default CI trace suppression remains intact. Only this job's explicit opt-in plus
+a newly generated authenticated synthetic fixture marker permits failure traces.
+The existing structured sanitizer runs before artifacts are uploaded; invalid
+ownership, failed sanitization and missing tooling purge sensitive artifacts.
+Unsupported or malformed sensitive records cause the affected archive to be removed
+fail-closed. An independent fail-closed verification step (.github/scripts/verify_sanitized_artifacts.py)
+runs after test suite execution or abnormal runner termination to prevent publishing
+unsanitized diagnostics while preserving safe JUnit reporting. The verifier returns a nonzero
+exit code upon any inspection or deletion failure, publishable XML is restricted to recognized
+JUnit reports with sensitive tokens redacted, and workflow upload steps execute only when
+independent verification succeeds (steps.<id>.outcome == 'success').
+Never upload the fixture directory, database or browser storage state.
+
+The stroke budget now excludes measured synthetic event-delivery pacing and also
+bounds actual Chromium main-thread work at the original5s. All240 samples and
+pixel/thumbnail/undo/redo assertions remain. Per-case timing attachments expose
+both clocks rather than treating driver frame cadence as application work.
+
+Three Bombadil properties remain opt-in: they require a Bombadil/browser scenario
+and bounded fuzz campaign; no authoritative execution is claimed. The two model
+smoke scripts require supplied model endpoints/models and remain explicit live
+operations. No credentials or live-service authorization are implied by CI setup.
+Firefox/WebKit alternatives remain opt-in and are not certified by Chromium CI.

mcp_servers/image_gen_server.py

@@ -33,7 +33,7 @@ async def list_tools() -> list[Tool]:
                     "prompt": {"type": "string", "description": "Image description prompt"},
                     "model": {"type": "string", "description": "Model name (auto-detects if omitted)"},
                     "size": {"type": "string", "description": "Image size (default 1024x1024)"},
-                    "quality": {"type": "string", "description": "Quality: low, medium, high, auto (default medium)"},
+                    "quality": {"type": "string", "description": "Quality: low, medium, high, auto (default configured quality, otherwise medium)"},
                 },
                 "required": ["prompt"],
             },
@@ -66,7 +66,7 @@ async def call_tool(name: str, arguments: dict) -> list[TextContent]:
 
         if not model_spec:
             model_spec = _settings.get("image_model", "")
-        if quality == "medium" and _settings.get("image_quality"):
+        if "quality" not in arguments and _settings.get("image_quality"):
             quality = _settings["image_quality"]
 
         # Auto-detect best available image model

package.json

@@ -9,7 +9,11 @@
     "test:photo-editor": "playwright test --config tests/e2e/playwright.config.js",
     "test:photo-editor:install": "playwright install chromium firefox webkit",
     "test:photo-editor:firefox": "PHOTO_EDITOR_E2E_BROWSER=firefox playwright test --config tests/e2e/playwright.config.js",
-    "test:photo-editor:webkit": "PHOTO_EDITOR_E2E_BROWSER=webkit playwright test --config tests/e2e/playwright.config.js"
+    "test:photo-editor:webkit": "PHOTO_EDITOR_E2E_BROWSER=webkit playwright test --config tests/e2e/playwright.config.js",
+    "test:ios-layout": "playwright test --config tests/e2e/ios-layout.config.cjs",
+    "test:ios-layout:chromium": "ODYSSEUS_LAYOUT_BROWSER=chromium playwright test --config tests/e2e/ios-layout.config.cjs",
+    "test:ios-layout:webkit": "ODYSSEUS_LAYOUT_BROWSER=webkit playwright test --config tests/e2e/ios-layout.config.cjs",
+    "test:ios-layout:install": "playwright install chromium webkit"
   },
   "devDependencies": {
     "@antithesishq/bombadil": "^0.7.0",

routes/auth_routes.py

@@ -887,6 +887,8 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
         """Create a new integration (admin only)."""
         _require_admin_user(request)
         body = await request.json()
+        if not isinstance(body, dict):
+            raise HTTPException(400, "Request body must be a JSON object")
         item = add_integration(body)
         return {"ok": True, "integration": mask_integration_secret(item)}
 
@@ -895,6 +897,8 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
         """Update an existing integration (admin only)."""
         _require_admin_user(request)
         body = await request.json()
+        if not isinstance(body, dict):
+            raise HTTPException(400, "Request body must be a JSON object")
         item = update_integration(integration_id, body)
         if not item:
             raise HTTPException(404, "Integration not found")

routes/calendar_routes.py

@@ -80,6 +80,18 @@ def _require_user(request: Request) -> str:
     return FALLBACK_OWNER
 
 
+async def _json_object_body(request: Request) -> dict:
+    """Parse a JSON object body. An unreadable body counts as empty; a body
+    that is valid JSON but not an object (a list, string or number) is a 400."""
+    try:
+        body = await request.json()
+    except Exception:
+        return {}
+    if not isinstance(body, dict):
+        raise HTTPException(400, "Request body must be a JSON object")
+    return body
+
+
 def _get_or_404_calendar(db, cal_id: str, owner: str) -> CalendarCal:
     cal = db.query(CalendarCal).filter(CalendarCal.id == cal_id).first()
     if not cal:
@@ -994,10 +1006,7 @@ def setup_calendar_routes(upload_handler=None) -> APIRouter:
     async def save_config(request: Request):
         """Legacy single-account endpoint — upserts the first account."""
         owner = _require_user(request)
-        try:
+        body = await _json_object_body(request)
-            body = await request.json()
-        except Exception:
-            body = {}
         accounts = _get_caldav_accounts(owner)
         if not (body.get("url") or "").strip():
             _save_caldav_accounts(owner, [])
@@ -1052,10 +1061,7 @@ def setup_calendar_routes(upload_handler=None) -> APIRouter:
         """Add a new CalDAV account."""
         import uuid as _uuid
         owner = _require_user(request)
-        try:
+        body = await _json_object_body(request)
-            body = await request.json()
-        except Exception:
-            body = {}
         from src.caldav_sync import validate_caldav_url
         try:
             url = validate_caldav_url(body.get("url", ""))
@@ -1080,10 +1086,7 @@ def setup_calendar_routes(upload_handler=None) -> APIRouter:
     async def update_caldav_account(account_id: str, request: Request):
         """Update an existing CalDAV account by id."""
         owner = _require_user(request)
-        try:
+        body = await _json_object_body(request)
-            body = await request.json()
-        except Exception:
-            body = {}
         accounts = _get_caldav_accounts(owner)
         idx = next((i for i, a in enumerate(accounts) if a.get("id") == account_id), None)
         if idx is None:
@@ -1124,10 +1127,7 @@ def setup_calendar_routes(upload_handler=None) -> APIRouter:
         test an already-saved account. Falls back to the first saved account
         when nothing is provided."""
         owner = _require_user(request)
-        try:
+        body = await _json_object_body(request)
-            body = await request.json()
-        except Exception:
-            body = {}
         url = (body.get("url") or "").strip()
         user = (body.get("username") or "").strip()
         pw = body.get("password") or ""

routes/cookbook_helpers.py

@@ -14,6 +14,7 @@ from fastapi import HTTPException
 from pydantic import BaseModel
 
 from routes._validators import validate_remote_host, validate_ssh_port
+from routes.cookbook_output import HF_PARTIAL_BLOB_HELPER
 from core.platform_compat import _ssh_exec_argv
 
 logger = logging.getLogger(__name__)
@@ -378,6 +379,7 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache:
     """
     lines = [
         "import json, os, re, shutil, subprocess, urllib.request",
+        *HF_PARTIAL_BLOB_HELPER.splitlines(),
         "models = []",
         "seen = set()",
         "BLOCKED_ROOTS = ('/sys', '/proc', '/dev', '/run', '/var/run')",
@@ -436,8 +438,10 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache:
         "        sz, nf, ic = 0, 0, False",
         "        if os.path.isdir(blobs):",
         "            for f in os.scandir(blobs):",
+        "                if f.name.endswith('.incomplete'):",
+        "                    if not is_partial_blob(f.path): continue",
+        "                    ic = True",
         "                if f.is_file(): nf += 1; sz += f.stat().st_size",
-        "                if f.name.endswith('.incomplete'): ic = True",
         "        snap = os.path.join(cache, d, 'snapshots')",
         "        def snapshot_size():",
         "            total, count, incomplete = 0, 0, False",
@@ -1030,6 +1034,166 @@ def _append_llama_cpp_linux_accel_build_lines(runner_lines: list[str]) -> None:
     runner_lines.append('  fi  # end _odysseus_have_prebuilt guard')
 
 
+ODYSSEUS_LLAMA_SHIM_MARKER = "# Auto-generated by Odysseus Cookbook: a `llama-server` lookalike"
+
+
+def _append_llama_server_resolution_definition_lines(runner_lines: list[str]) -> None:
+    """Define the helper shell function that discovers, reconciles and prioritizes llama-server."""
+    marker = f"'{_bash_squote(ODYSSEUS_LLAMA_SHIM_MARKER)}'"
+    runner_lines.append('_odysseus_resolve_llama_server() {')
+    runner_lines.append('  _odysseus_llama_bin=""')
+    runner_lines.append(f'  local _marker={marker}')
+    runner_lines.append('  local _saved_ifs="$IFS"')
+    runner_lines.append("  IFS=':'")
+    runner_lines.append('  local -a _parts')
+    runner_lines.append('  read -r -a _parts <<< "$PATH"')
+    runner_lines.append('  IFS="$_saved_ifs"')
+    runner_lines.append('  local _dir _cand')
+    runner_lines.append('  for _dir in "${_parts[@]}"; do')
+    runner_lines.append('    [ -n "$_dir" ] || continue')
+    runner_lines.append('    _cand="$_dir/llama-server"')
+    runner_lines.append('    [ -x "$_cand" ] || continue')
+    runner_lines.append('    if [ -f "$_cand" ] && [ ! -L "$_cand" ] && grep -Fxq "$_marker" "$_cand" 2>/dev/null; then')
+    runner_lines.append('      continue')
+    runner_lines.append('    fi')
+    runner_lines.append('    if "$_cand" --version >/dev/null 2>&1; then')
+    runner_lines.append('      _odysseus_llama_bin="$_cand"')
+    runner_lines.append('      break')
+    runner_lines.append('    fi')
+    runner_lines.append('  done')
+    runner_lines.append('  if [ -f "$HOME/bin/llama-server" ] && [ ! -L "$HOME/bin/llama-server" ] && grep -Fxq "$_marker" "$HOME/bin/llama-server" 2>/dev/null; then')
+    runner_lines.append('    if [ -n "$_odysseus_llama_bin" ]; then')
+    runner_lines.append('      echo "[odysseus] Native llama-server detected — retiring Odysseus compatibility shim in ~/bin"')
+    runner_lines.append('      rm -f "$HOME/bin/llama-server"')
+    runner_lines.append('      hash -r')
+    runner_lines.append('    elif python3 -c "import llama_cpp.server" 2>/dev/null; then')
+    runner_lines.append('      _odysseus_llama_bin="$HOME/bin/llama-server"')
+    runner_lines.append('    else')
+    runner_lines.append('      echo "[odysseus] Removing stale Odysseus llama-server shim (python3 -m llama_cpp.server unavailable)"')
+    runner_lines.append('      rm -f "$HOME/bin/llama-server"')
+    runner_lines.append('      hash -r')
+    runner_lines.append('    fi')
+    runner_lines.append('  fi')
+    runner_lines.append('  if [ -n "$_odysseus_llama_bin" ]; then')
+    runner_lines.append('    local _cand_dir')
+    runner_lines.append('    _cand_dir="$(dirname "$_odysseus_llama_bin")"')
+    runner_lines.append('    if [ "${PATH%%:*}" != "$_cand_dir" ]; then')
+    runner_lines.append('      export PATH="$_cand_dir:$PATH"')
+    runner_lines.append('    fi')
+    runner_lines.append('    hash -r')
+    runner_lines.append('    return 0')
+    runner_lines.append('  fi')
+    runner_lines.append('  return 1')
+    runner_lines.append('}')
+
+
+def _append_llama_server_shim_definition_lines(runner_lines: list[str]) -> None:
+    """Define the helper shell function that generates the Odysseus compatibility shim."""
+    _append_llama_server_resolution_definition_lines(runner_lines)
+    marker = f"'{_bash_squote(ODYSSEUS_LLAMA_SHIM_MARKER)}'"
+    runner_lines.append('_odysseus_create_llama_shim() {')
+    runner_lines.append('  local _target="$HOME/bin/llama-server"')
+    runner_lines.append(f'  local _marker={marker}')
+    runner_lines.append('  if [ -e "$_target" ] || [ -L "$_target" ]; then')
+    runner_lines.append('    if [ -f "$_target" ] && [ ! -L "$_target" ] && grep -Fxq "$_marker" "$_target" 2>/dev/null; then')
+    runner_lines.append('      : # existing Odysseus shim — safe to replace')
+    runner_lines.append('    else')
+    runner_lines.append('      echo "[odysseus] WARNING: $_target already exists and is not an Odysseus-managed shim; leaving untouched." >&2')
+    runner_lines.append('      return 1')
+    runner_lines.append('    fi')
+    runner_lines.append('  fi')
+    runner_lines.append('  mkdir -p "$HOME/bin" || return 1')
+    runner_lines.append('  local _tmp')
+    runner_lines.append('  _tmp="$(mktemp "$HOME/bin/.llama-server.odysseus.tmp.XXXXXX")" || {')
+    runner_lines.append('    echo "[odysseus] ERROR: unable to create secure temporary llama-server shim in $HOME/bin" >&2')
+    runner_lines.append('    return 1')
+    runner_lines.append('  }')
+    runner_lines.append("  if ! cat > \"$_tmp\" <<'_ODY_LLAMA_SHIM_EOF'")
+    runner_lines.append('#!/usr/bin/env bash')
+    runner_lines.append(ODYSSEUS_LLAMA_SHIM_MARKER)
+    runner_lines.append('# that translates the native CLI to `python -m llama_cpp.server`.')
+    runner_lines.append('# Lets cookbook-generated launch commands run unchanged on hosts')
+    runner_lines.append('# where only the pip llama-cpp-python package is installed.')
+    runner_lines.append('ARGS=()')
+    runner_lines.append('while [ $# -gt 0 ]; do')
+    runner_lines.append('  case "$1" in')
+    runner_lines.append('    -ngl|--gpu-layers|--n-gpu-layers) ARGS+=(--n_gpu_layers "$2"); shift 2 ;;')
+    runner_lines.append('    -c|--ctx-size) ARGS+=(--n_ctx "$2"); shift 2 ;;')
+    runner_lines.append('    -b|--batch-size) ARGS+=(--n_batch "$2"); shift 2 ;;')
+    runner_lines.append('    -ub|--ubatch-size) shift 2 ;;  # llama-cpp-python has no separate ubatch')
+    runner_lines.append('    --flash-attn) ARGS+=(--flash_attn true); shift 2 ;;')
+    runner_lines.append('    --cache-type-k) ARGS+=(--type_k "$2"); shift 2 ;;')
+    runner_lines.append('    --cache-type-v) ARGS+=(--type_v "$2"); shift 2 ;;')
+    runner_lines.append('    --n-cpu-moe) ARGS+=(--n_cpu_moe "$2"); shift 2 ;;')
+    runner_lines.append('    --mmproj) ARGS+=(--clip_model_path "$2"); shift 2 ;;')
+    runner_lines.append('    --image-max-tokens) shift 2 ;;  # native-only')
+    runner_lines.append('    --no-mmap) ARGS+=(--no_mmap true); shift ;;')
+    runner_lines.append('    --no-warmup) shift ;;  # native-only')
+    runner_lines.append('    --chat-template) ARGS+=(--chat_format "$2"); shift 2 ;;')
+    runner_lines.append('    --fit|--split-mode|--tensor-split|--main-gpu|--parallel) shift 2 ;;  # native-only')
+    runner_lines.append('    --mlock) ARGS+=(--use_mlock true); shift ;;')
+    runner_lines.append('    *) ARGS+=("$1"); shift ;;')
+    runner_lines.append('  esac')
+    runner_lines.append('done')
+    runner_lines.append('exec python3 -m llama_cpp.server "${ARGS[@]}"')
+    runner_lines.append('_ODY_LLAMA_SHIM_EOF')
+    runner_lines.append('  then')
+    runner_lines.append('    rm -f -- "$_tmp"')
+    runner_lines.append('    return 1')
+    runner_lines.append('  fi')
+    runner_lines.append('  chmod +x "$_tmp" || { rm -f -- "$_tmp"; return 1; }')
+    runner_lines.append('  mv -f -- "$_tmp" "$_target" || { rm -f -- "$_tmp"; return 1; }')
+    runner_lines.append('  hash -r')
+    runner_lines.append('  echo "[odysseus] Created llama-server shim → python -m llama_cpp.server (no native binary needed)"')
+    runner_lines.append('}')
+
+
+def _append_llama_server_shim_reconciliation_lines(runner_lines: list[str]) -> None:
+    """Reconcile an existing ~/bin/llama-server if it is owned by Odysseus Cookbook.
+
+    Invariants:
+    - Never touch an arbitrary user-owned executable or symlink.
+    - If ~/bin/llama-server is an Odysseus generated shim:
+        - If a native llama-server is present and healthy elsewhere on PATH, remove the shim
+          so the native binary is used.
+        - If python3 -m llama_cpp.server is unavailable, remove the stale shim
+          and hash -r so the fallback/build path is triggered.
+        - If no healthy native server exists and llama_cpp.server is working, preserve it.
+    - If a healthy native candidate exists on PATH (even later in PATH), prioritize it and
+      ensure the launched executable is the validated candidate.
+    """
+    runner_lines.append('[ -r ~/.config/odysseus-llama-cpp-env ] && . ~/.config/odysseus-llama-cpp-env')
+    runner_lines.append('if ! declare -F _odysseus_resolve_llama_server >/dev/null 2>&1; then')
+    _append_llama_server_resolution_definition_lines(runner_lines)
+    runner_lines.append('fi')
+    runner_lines.append('_odysseus_resolve_llama_server || true')
+
+
+def _append_llama_server_finalization_lines(runner_lines: list[str], platform_hint: str = "") -> None:
+    """Reconcile post-install launcher availability and enforce fail-closed invariant.
+
+    Guarantees that if python3 -m llama_cpp.server is usable but the `llama-server`
+    command is unresolvable, the compatibility launcher is created. If no usable
+    launcher exists after all build/install attempts, sets ODYSSEUS_PREFLIGHT_EXIT=127.
+    """
+    err_msg = (
+        f"ERROR: llama.cpp serving is not available on {platform_hint} after install attempts."
+        if platform_hint
+        else "ERROR: llama.cpp serving is not available after install/build attempts."
+    )
+    runner_lines.append('if ! declare -F _odysseus_resolve_llama_server >/dev/null 2>&1; then')
+    _append_llama_server_resolution_definition_lines(runner_lines)
+    runner_lines.append('fi')
+    runner_lines.append('if [ -z "$_odysseus_llama_bin" ] && python3 -c "import llama_cpp.server" 2>/dev/null; then')
+    runner_lines.append('  _odysseus_create_llama_shim')
+    runner_lines.append('  _odysseus_resolve_llama_server || true')
+    runner_lines.append('fi')
+    runner_lines.append('if [ -z "$_odysseus_llama_bin" ]; then')
+    runner_lines.append(f'  echo "{err_msg}"')
+    runner_lines.append('  ODYSSEUS_PREFLIGHT_EXIT=127')
+    runner_lines.append('fi')
+
+
 def _llama_cpp_rebuild_cmd(update_source: bool = False) -> str:
     """Shell command that clears the Cookbook-managed llama.cpp build.
 

routes/cookbook_output.py

@@ -8,6 +8,20 @@ import re
 
 _FETCHING_ZERO_FILES_RE = re.compile(r"Fetching\s+0\s+files", re.IGNORECASE)
 
+# Shared source embedded in local/remote inventory and status probes. A blank
+# retry marker beside its completed, nonempty content-addressed blob is stale;
+# a blank marker alone may be a download that has not received its first byte.
+HF_PARTIAL_BLOB_HELPER = """def is_partial_blob(path):
+    if not path.endswith('.incomplete'):
+        return False
+    try:
+        completed = path[:-len('.incomplete')]
+        return (os.path.getsize(path) != 0 or not os.path.isfile(completed)
+                or os.path.getsize(completed) == 0)
+    except OSError:
+        return True
+"""
+
 # Probe scripts for the dead-session download check, run as
 # `python3 -c <PROBE> <repo_id> <cache_root>` (locally or over SSH).
 # cache_root is the task's custom download dir, '' for the default HF cache.
@@ -15,7 +29,7 @@ _FETCHING_ZERO_FILES_RE = re.compile(r"Fetching\s+0\s+files", re.IGNORECASE)
 # HF_HOME=<local_dir>, so that task's cache lives under <local_dir>/hub, and
 # the probe process's own environment knows nothing about it.
 HF_CACHE_COMPLETE_PROBE = (
-    "import os,sys;"
+    "import os,sys\n" + HF_PARTIAL_BLOB_HELPER +
     "repo=sys.argv[1];"
     "root=os.path.expanduser(sys.argv[2]) if len(sys.argv)>2 and sys.argv[2] else '';"
     "base=os.path.join(root,'hub') if root else (os.environ.get('HUGGINGFACE_HUB_CACHE') or os.path.join(os.environ.get('HF_HOME', os.path.expanduser('~/.cache/huggingface')), 'hub'));"
@@ -24,18 +38,18 @@ HF_CACHE_COMPLETE_PROBE = (
     "ok=os.path.isdir(snap) and any(os.path.isdir(os.path.join(snap,x)) and os.listdir(os.path.join(snap,x)) for x in os.listdir(snap));"
     "inc=False;"
     "blobs=os.path.join(d,'blobs');"
-    "inc=os.path.isdir(blobs) and any(x.endswith('.incomplete') for x in os.listdir(blobs));"
+    "inc=os.path.isdir(blobs) and any(is_partial_blob(os.path.join(blobs,x)) for x in os.listdir(blobs));"
     "sys.exit(0 if ok and not inc else 1)"
 )
 
 HF_CACHE_INCOMPLETE_PROBE = (
-    "import os,sys;"
+    "import os,sys\n" + HF_PARTIAL_BLOB_HELPER +
     "repo=sys.argv[1];"
     "root=os.path.expanduser(sys.argv[2]) if len(sys.argv)>2 and sys.argv[2] else '';"
     "base=os.path.join(root,'hub') if root else (os.environ.get('HUGGINGFACE_HUB_CACHE') or os.path.join(os.environ.get('HF_HOME', os.path.expanduser('~/.cache/huggingface')), 'hub'));"
     "d=os.path.join(base,'models--'+repo.replace('/','--'));"
     "blobs=os.path.join(d,'blobs');"
-    "inc=os.path.isdir(blobs) and any(x.endswith('.incomplete') for x in os.listdir(blobs));"
+    "inc=os.path.isdir(blobs) and any(is_partial_blob(os.path.join(blobs,x)) for x in os.listdir(blobs));"
     "sys.exit(0 if inc else 1)"
 )
 

routes/cookbook_routes.py

@@ -57,9 +57,11 @@ from routes.cookbook_helpers import (
     _pip_install_no_cache, _user_shell_path_bootstrap, _venv_safe_local_pip_install_cmd,
     _diagnose_serve_output, run_ssh_command_async,
     _ollama_bind_from_cmd, _pip_install_fallback_chain, _pip_install_no_cache,
-    _user_shell_path_bootstrap, _venv_safe_local_pip_install_cmd,
     _append_pip_install_runner_lines, _pip_install_command_without_break_system_packages,
     _normalize_llama_cpp_python_cache_types,
+    _append_llama_server_shim_definition_lines,
+    _append_llama_server_shim_reconciliation_lines,
+    _append_llama_server_finalization_lines,
     ModelDownloadRequest, ServeRequest,
 )
 
@@ -2161,7 +2163,7 @@ def setup_cookbook_routes() -> APIRouter:
                 ps_lines.append('}')
             elif "llama_cpp" in req.cmd or "llama-server" in req.cmd:
                 ps_lines.append('# Auto-install llama-cpp-python if missing')
-                ps_lines.append('try { python -c "import llama_cpp" 2>$null } catch {}')
+                ps_lines.append('try { python -c "import llama_cpp.server" 2>$null } catch {}')
                 ps_lines.append('if ($LASTEXITCODE -ne 0) {')
                 ps_lines.append('  Write-Host "Installing llama-cpp-python..."')
                 ps_lines.append('  python -m pip install llama-cpp-python[server]')
@@ -2247,14 +2249,17 @@ def setup_cookbook_routes() -> APIRouter:
                 # ollama is found (otherwise macOS falls back to a slow source build).
                 # /opt/homebrew = Apple Silicon, /usr/local = Intel; harmless on Linux.
                 runner_lines.append('export PATH="$HOME/.local/bin:$HOME/bin:$HOME/llama.cpp/build/bin:/opt/homebrew/bin:/usr/local/bin:$PATH"')
+                _append_llama_server_shim_definition_lines(runner_lines)
+                _append_llama_server_shim_reconciliation_lines(runner_lines)
                 runner_lines.append('if [ -d /data/data/com.termux ]; then')
                 runner_lines.append('  # Termux: no native build — use the Python bindings (CPU).')
-                runner_lines.append('  if ! python3 -c "import llama_cpp" 2>/dev/null; then')
+                runner_lines.append('  if ! python3 -c "import llama_cpp.server" 2>/dev/null; then')
                 runner_lines.append('    pkg install -y cmake 2>/dev/null')
                 runner_lines.append('    pip install numpy diskcache jinja2 2>/dev/null')
                 runner_lines.append('    CMAKE_ARGS="-DGGML_BLAS=OFF -DGGML_LLAMAFILE=OFF" pip install \'llama-cpp-python[server]\' --no-build-isolation --no-cache-dir 2>&1 || true')
                 runner_lines.append('  fi')
-                runner_lines.append('elif ! command -v llama-server &>/dev/null; then')
+                _append_llama_server_finalization_lines(runner_lines, platform_hint="Termux")
+                runner_lines.append('elif [ -z "$_odysseus_llama_bin" ]; then')
                 runner_lines.append('  echo "Native llama-server not found — building from source (one-time, may take a few minutes)..."')
                 runner_lines.append('  mkdir -p ~/bin')
                 runner_lines.append('  cd ~ && [ -d llama.cpp ] || git clone --depth 1 https://github.com/ggml-org/llama.cpp')
@@ -2271,8 +2276,7 @@ def setup_cookbook_routes() -> APIRouter:
                 # would reuse the bad settings and fail again. CMAKE_BUILD_TYPE is
                 # explicit so the binary is optimized (Metal auto-enables on macOS).
                 runner_lines.append('    cd ~/llama.cpp && rm -rf build && cmake -B build -DCMAKE_BUILD_TYPE=Release \\')
-                runner_lines.append('      && cmake --build build -j"$NPROC" --target llama-server \\')
+                runner_lines.append('      && cmake --build build -j"$NPROC" --target llama-server')
-                runner_lines.append('      && ln -sf ~/llama.cpp/build/bin/llama-server ~/bin/llama-server')
                 runner_lines.append('  else')
                 _append_llama_cpp_linux_accel_build_lines(runner_lines)
                 runner_lines.append('  fi')
@@ -2296,58 +2300,13 @@ def setup_cookbook_routes() -> APIRouter:
                 runner_lines.append('      fi')
                 runner_lines.append('    fi')
                 runner_lines.append('  fi')
-                # SHORT-CIRCUIT before the build/pip fallback: if the
+                runner_lines.append('  _odysseus_resolve_llama_server || true')
-                # native binary is missing but llama_cpp Python is already
+                # If the native build failed, fall back to the Python bindings.
-                # installed, drop a wrapper at ~/bin/llama-server that
+                runner_lines.append('  if [ -z "$_odysseus_llama_bin" ] && ! python3 -c "import llama_cpp.server" 2>/dev/null; then')
-                # translates llama-server CLI args to llama_cpp.server's
-                # underscore-style flags. The user's serve command stays
-                # `llama-server ...` and "just works" — no build, no cmake,
-                # no second install. This is the path that unblocks every
-                # remote where pip-installed llama-cpp-python is already
-                # working but Cookbook used to insist on a native binary.
-                runner_lines.append('  if ! command -v llama-server >/dev/null 2>&1 && python3 -c "import llama_cpp" 2>/dev/null; then')
-                runner_lines.append('    mkdir -p ~/bin')
-                runner_lines.append('    cat > ~/bin/llama-server <<\'_ODY_LLAMA_SHIM_EOF\'')
-                runner_lines.append('#!/usr/bin/env bash')
-                runner_lines.append('# Auto-generated by Odysseus Cookbook: a `llama-server` lookalike')
-                runner_lines.append('# that translates the native CLI to `python -m llama_cpp.server`.')
-                runner_lines.append('# Lets cookbook-generated launch commands run unchanged on hosts')
-                runner_lines.append('# where only the pip llama-cpp-python package is installed.')
-                runner_lines.append('ARGS=()')
-                runner_lines.append('while [ $# -gt 0 ]; do')
-                runner_lines.append('  case "$1" in')
-                runner_lines.append('    -ngl|--gpu-layers|--n-gpu-layers) ARGS+=(--n_gpu_layers "$2"); shift 2 ;;')
-                runner_lines.append('    -c|--ctx-size) ARGS+=(--n_ctx "$2"); shift 2 ;;')
-                runner_lines.append('    -b|--batch-size) ARGS+=(--n_batch "$2"); shift 2 ;;')
-                runner_lines.append('    -ub|--ubatch-size) shift 2 ;;  # llama-cpp-python has no separate ubatch')
-                runner_lines.append('    --flash-attn) ARGS+=(--flash_attn true); shift 2 ;;')
-                runner_lines.append('    --cache-type-k) ARGS+=(--type_k "$2"); shift 2 ;;')
-                runner_lines.append('    --cache-type-v) ARGS+=(--type_v "$2"); shift 2 ;;')
-                runner_lines.append('    --n-cpu-moe) ARGS+=(--n_cpu_moe "$2"); shift 2 ;;')
-                runner_lines.append('    --mmproj) ARGS+=(--clip_model_path "$2"); shift 2 ;;')
-                runner_lines.append('    --image-max-tokens) shift 2 ;;  # native-only')
-                runner_lines.append('    --no-mmap) ARGS+=(--no_mmap true); shift ;;')
-                runner_lines.append('    --no-warmup) shift ;;  # native-only')
-                runner_lines.append('    --chat-template) ARGS+=(--chat_format "$2"); shift 2 ;;')
-                runner_lines.append('    --fit|--split-mode|--tensor-split|--main-gpu|--parallel) shift 2 ;;  # native-only')
-                runner_lines.append('    --mlock) ARGS+=(--use_mlock true); shift ;;')
-                runner_lines.append('    *) ARGS+=("$1"); shift ;;')
-                runner_lines.append('  esac')
-                runner_lines.append('done')
-                runner_lines.append('exec python3 -m llama_cpp.server "${ARGS[@]}"')
-                runner_lines.append('_ODY_LLAMA_SHIM_EOF')
-                runner_lines.append('    chmod +x ~/bin/llama-server')
-                runner_lines.append('    echo "[odysseus] Created llama-server shim → python -m llama_cpp.server (no native binary needed)"')
-                runner_lines.append('  fi')
-                runner_lines.append('  # If the native build failed, fall back to the Python bindings.')
-                runner_lines.append('  if ! command -v llama-server &>/dev/null && ! python3 -c "import llama_cpp" 2>/dev/null; then')
                 runner_lines.append('    echo "llama-server build failed — installing Python bindings as fallback..."')
                 runner_lines.append(f"    {_pip_install_fallback_chain('llama-cpp-python[server]', python_cmd='pip')} || true")
                 runner_lines.append('  fi')
-                runner_lines.append('  if ! command -v llama-server &>/dev/null && ! python3 -c "import llama_cpp" 2>/dev/null; then')
+                _append_llama_server_finalization_lines(runner_lines)
-                runner_lines.append('    echo "ERROR: llama.cpp serving is not available after install/build attempts."')
-                runner_lines.append('    ODYSSEUS_PREFLIGHT_EXIT=127')
-                runner_lines.append('  fi')
                 runner_lines.append('fi')
             elif re.search(r"\bollama\s+serve\b", req.cmd):
                 handled_ollama_serve = True
@@ -4343,7 +4302,7 @@ def setup_cookbook_routes() -> APIRouter:
             tmux output can stop at a stale progress line if the pane/session
             disappears before Cookbook captures the final DOWNLOAD_OK marker.
             In that case, trust the cache shape: a snapshot directory with files
-            and no *.incomplete blobs means HuggingFace finished materializing the
+            and no unfinished blobs means HuggingFace finished materializing the
             model. cache_root is the task's custom download dir — the runner
             pointed HF_HOME there, so the cache lives under <cache_root>/hub,
             not wherever this probe's environment says.
@@ -4368,7 +4327,8 @@ def setup_cookbook_routes() -> APIRouter:
             """Best-effort check for resumable HF partial blobs.
 
             A lost SSH/tmux session can leave a real download still incomplete.
-            Treat any *.incomplete blob as stronger evidence than stale
+            Treat unfinished blobs (excluding empty markers paired with a
+            completed blob) as stronger evidence than stale
             "100%" lines in the captured pane output.
             """
             if not repo_id or "/" not in repo_id:

routes/document/document_routes.py

@@ -781,9 +781,11 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         except Exception as e:
             logger.warning("Failed to parse export request body, defaulting to empty", exc_info=e)
             data = {}
-        ids = data.get("ids") or []
+        ids = data.get("ids") if isinstance(data, dict) else None
         if not ids:
             raise HTTPException(400, "No documents specified")
+        if not isinstance(ids, list) or not all(isinstance(i, str) for i in ids):
+            raise HTTPException(400, "ids must be a list of document ids")
         _ext = {
             "javascript": ".js", "python": ".py", "html": ".html", "css": ".css",
             "markdown": ".md", "json": ".json", "yaml": ".yml", "bash": ".sh",

routes/font_routes.py

@@ -19,9 +19,10 @@ def _split_family_token(token):
 def _derive_family(filename):
     """Derive a font-family name from a filename like 'JetBrainsMono-Regular.woff2' → 'JetBrains Mono'."""
     name = os.path.splitext(filename)[0]
-    # Strip common weight/style suffixes
+    # Strip weight/style suffixes, including BoldItalic and Bold-Italic.
     name = re.sub(
-        r'[-_ ]?(Thin|ExtraLight|UltraLight|Light|Regular|Medium|SemiBold|DemiBold|Bold|ExtraBold|UltraBold|Black|Heavy|Italic|Oblique|Variable|VF)$',
+        r'[-_ ]?(Thin|ExtraLight|UltraLight|Light|Regular|Medium|SemiBold|DemiBold|Bold|ExtraBold|UltraBold|Black|Heavy|Italic|Oblique|Variable|VF)'
+        r'(?:[-_ ]?(?:Italic|Oblique))?$',
         '', name, flags=re.IGNORECASE
     )
     # Replace dashes/underscores with spaces

routes/gallery/gallery_routes.py

@@ -2433,11 +2433,10 @@ def setup_gallery_routes() -> APIRouter:
         image_b64 = body.get("image")
         amount = body.get("amount", 50) / 100.0
 
-        from PIL import Image, ImageFilter
+        from PIL import ImageFilter
         import base64, io
 
-        img_bytes = _b64_bytes(image_b64)
+        img = _b64_to_pil_image(image_b64, mode="RGB")
-        img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
 
         # Unsharp mask: radius=2, percent=amount*200, threshold=3
         sharpened = img.filter(ImageFilter.UnsharpMask(radius=2, percent=int(amount * 200), threshold=3))
@@ -2472,8 +2471,7 @@ def setup_gallery_routes() -> APIRouter:
         except ImportError:
             raise HTTPException(500, "Server missing a required dependency")
         # Decode source image (RGB; Real-ESRGAN doesn't preserve alpha).
-        img_bytes = _b64_bytes(image_b64)
+        src = _b64_to_pil_image(image_b64, mode="RGB")
-        src = Image.open(io.BytesIO(img_bytes)).convert("RGB")
         try:
             patch_realesrgan_torchvision_compat()
             from realesrgan import RealESRGANer
@@ -2525,8 +2523,7 @@ def setup_gallery_routes() -> APIRouter:
             import numpy as np
         except ImportError:
             raise HTTPException(500, "Server missing a required dependency")
-        img_bytes = _b64_bytes(image_b64)
+        src = _b64_to_pil_image(image_b64, mode="RGB")
-        src = Image.open(io.BytesIO(img_bytes)).convert("RGB")
         try:
             patch_realesrgan_torchvision_compat()
             from basicsr.archs.rrdbnet_arch import RRDBNet
@@ -2778,8 +2775,7 @@ def setup_gallery_routes() -> APIRouter:
         from PIL import Image
         import base64, io
 
-        img_bytes = _b64_bytes(image_b64)
+        img = _b64_to_pil_image(image_b64)
-        img = Image.open(io.BytesIO(img_bytes)).convert("RGBA")
         W, H = img.size
 
         hint = None
@@ -2868,8 +2864,7 @@ def setup_gallery_routes() -> APIRouter:
         from PIL import Image, ImageFilter, ImageEnhance
         import numpy as np
 
-        img_bytes = _b64_bytes(image_b64)
+        img = _b64_to_pil_image(image_b64, mode="RGB")
-        img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
 
         # Try GFPGAN first (AI face restoration)
         try:

routes/note/note_routes.py

@@ -728,32 +728,35 @@ def setup_note_routes(task_scheduler=None, upload_handler=None):
             )
             if body.title is not None:
                 note.title = body.title
-            if body.content is not None:
+            # Omitted fields preserve their values; explicit null clears the
+            # nullable fields used by the editor's removal controls.
+            supplied = body.model_fields_set
+            if "content" in supplied:
                 note.content = body.content
-            if body.items is not None:
+            if "items" in supplied:
-                note.items = json.dumps(body.items)
+                note.items = json.dumps(body.items) if body.items is not None else None
                 flag_modified(note, "items")
             if body.note_type is not None:
                 note.note_type = body.note_type
-            if body.color is not None:
+            if "color" in supplied:
                 note.color = body.color
-            if body.label is not None:
+            if "label" in supplied:
                 note.label = body.label
             if body.pinned is not None:
                 note.pinned = body.pinned
             if body.archived is not None:
                 note.archived = body.archived
-            if body.due_date is not None:
+            if "due_date" in supplied:
                 note.due_date = body.due_date
-            if body.image_url is not None:
+            if "image_url" in supplied:
                 note.image_url = body.image_url
-            if body.gallery_id is not None:
+            if "gallery_id" in supplied:
                 note.gallery_id = body.gallery_id
             if body.repeat is not None:
                 note.repeat = body.repeat
             if body.sort_order is not None:
                 note.sort_order = body.sort_order
-            if body.agent_session_id is not None:
+            if "agent_session_id" in supplied:
                 note.agent_session_id = body.agent_session_id
 
             db.commit()

routes/session_routes.py

@@ -1315,7 +1315,7 @@ def setup_session_routes(
     def auto_sort_sessions(request: Request, skip_llm: bool = False):
         """Use AI to categorize all sessions into folders.
 
-        Phase 1 deletes empty/throwaway sessions and Phase 2 asks the LLM
+        Phase 1 deletes empty/Incognito sessions and Phase 2 asks the LLM
         to assign folders. When `skip_llm=true` the endpoint returns
         after Phase 1 — used by the "Tidy (no AI)" UI affordance so
         users can clean junk without spending tokens.
@@ -1326,21 +1326,11 @@ def setup_session_routes(
         single_user_mode = not user and _auth_disabled()
         user_sessions = session_manager.get_sessions_for_user(user)
 
-        # Delete empty and throwaway sessions before sorting
+        # Saved conversations are never disposable based on their title or prompt.
         from core.database import ChatMessage as DbMsg
         db = SessionLocal()
         deleted_empty = 0
         deleted_throwaway = 0
-        # Names that indicate a throwaway/test session (case-insensitive exact or prefix match)
-        _THROWAWAY_NAMES = {
-            "test", "testing", "asdf", "asd", "hello", "hi", "hey",
-            "yo", "sup", "hola", "hii", "hiii", "heyo",
-            "foo", "bar", "baz", "tmp", "temp", "scratch", "untitled",
-            "new chat", "delete", "remove", "junk", "trash", "xxx",
-            "abc", "qwerty", "blah", "stuff", "whatever", "idk",
-            "ok", "lol", "bruh", "hmm", "hm", "meh",
-        }
-        _THROWAWAY_MAX_MESSAGES = 4  # only delete if <= this many messages
         try:
             rows_q = db.query(DbSession).filter(DbSession.archived == False)
             if user:
@@ -1349,15 +1339,11 @@ def setup_session_routes(
                 rows_q = rows_q.filter(DbSession.owner == user)
             rows = rows_q.limit(2000).all()
             folder_map = {r.id: r.folder for r in rows}
-            # Precompute per-session message counts in TWO aggregate queries
+            # Precompute per-session message counts in one aggregate query
             # instead of 1–3 queries PER session — with many chats the per-row
             # loop was doing thousands of round-trips and blowing the timeout.
             from sqlalchemy import func as _sa_func
             _counts = dict(db.query(DbMsg.session_id, _sa_func.count(DbMsg.id)).group_by(DbMsg.session_id).all())
-            _asst_counts = dict(
-                db.query(DbMsg.session_id, _sa_func.count(DbMsg.id))
-                .filter(DbMsg.role == "assistant").group_by(DbMsg.session_id).all()
-            )
             cleanup_now = utcnow_naive()
             for row in rows:
                 # Never delete important sessions
@@ -1367,9 +1353,10 @@ def setup_session_routes(
                 if (row.name or "").strip() == "Incognito":
                     should_delete = True
                     deleted_throwaway += 1
-                    db.delete(row)
                     if hasattr(session_manager, 'delete_session'):
                         session_manager.delete_session(row.id)
+                    else:
+                        db.delete(row)
                     continue
                 if is_session_recently_active(row, now=cleanup_now):
                     continue
@@ -1378,31 +1365,11 @@ def setup_session_routes(
                 if msg_count == 0:
                     should_delete = True
                     deleted_empty += 1
-                elif msg_count <= _THROWAWAY_MAX_MESSAGES:
-                    name = (row.name or "").strip().lower()
-                    # Check first user message content (AI renames sessions, so
-                    # "hi" becomes "Casual Greeting Exchange" — name alone won't match)
-                    first_msg = db.query(DbMsg.content).filter(
-                        DbMsg.session_id == row.id, DbMsg.role == "user"
-                    ).order_by(DbMsg.timestamp).first()
-                    first_text = (first_msg[0] or "").strip().lower() if first_msg else ""
-                    # Count assistant messages — if user sent something but AI never replied, it's dead
-                    assistant_count = _asst_counts.get(row.id, 0)
-                    if name in _THROWAWAY_NAMES or name.startswith("chat:") or first_text in _THROWAWAY_NAMES:
-                        should_delete = True
-                        deleted_throwaway += 1
-                    # Single user message with no AI response = dead session
-                    elif msg_count == 1 and assistant_count == 0:
-                        should_delete = True
-                        deleted_throwaway += 1
-                    # Short phrase (1-3 words) with no real AI conversation (<=2 msgs)
-                    elif msg_count <= 2 and first_text and len(first_text.split()) <= 3 and len(first_text) <= 40:
-                        should_delete = True
-                        deleted_throwaway += 1
                 if should_delete:
-                    db.delete(row)
                     if hasattr(session_manager, 'delete_session'):
                         session_manager.delete_session(row.id)
+                    else:
+                        db.delete(row)
             if deleted_empty or deleted_throwaway:
                 db.commit()
                 logger.info(f"Auto-sort: deleted {deleted_empty} empty + {deleted_throwaway} throwaway sessions")

routes/task/task_routes.py

@@ -20,7 +20,7 @@ from src.task_action_policy import (
     is_admin_only_task_action,
     owner_has_admin_task_privileges,
 )
-from src.task_scheduler import compute_next_run, HOUSEKEEPING_DEFAULTS
+from src.task_scheduler import compute_next_run, is_valid_cron, HOUSEKEEPING_DEFAULTS
 from routes.prefs_routes import _load_for_user, _save_for_user
 
 logger = logging.getLogger(__name__)
@@ -475,10 +475,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
         if req.trigger_type == "schedule" and req.schedule == "cron" and not req.cron_expression:
             raise HTTPException(400, "Cron expression is required for cron schedule")
         if req.trigger_type == "schedule" and req.schedule == "cron" and req.cron_expression:
-            try:
+            if not is_valid_cron(req.cron_expression):
-                from croniter import croniter
-                croniter(req.cron_expression)
-            except Exception:
                 raise HTTPException(400, "Invalid cron expression")
         if req.trigger_type == "event" and not req.trigger_event:
             raise HTTPException(400, "Event name is required for event-triggered tasks")
@@ -788,10 +785,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 task.character_id = req.character_id or None
             if req.cron_expression is not None:
                 if req.cron_expression:
-                    try:
+                    if not is_valid_cron(req.cron_expression):
-                        from croniter import croniter
-                        croniter(req.cron_expression)
-                    except Exception:
                         raise HTTPException(400, "Invalid cron expression")
                 task.cron_expression = req.cron_expression or None
 

scripts/odysseus-gallery

@@ -6,6 +6,7 @@ that are awkward at the shell). Filters images by tag, prompt, favorite,
 album, and surfaces EXIF metadata.
 
     odysseus-gallery list --limit 20 --favorites
+    odysseus-gallery list --tag "sunset,beach"  # all tags, manual or AI
     odysseus-gallery show IMAGE_ID
     odysseus-gallery albums | jq -r '.[] | .name'
     odysseus-gallery search "sunset"          # matches prompt + tags
@@ -83,7 +84,13 @@ def cmd_list(args):
                 fail(f"no album named {args.album!r}")
             q = q.filter(GalleryImage.album_id == al.id)
         if args.tag:
-            q = q.filter(GalleryImage.tags.ilike(f"%{args.tag}%"))
+            for tag in (value.strip() for value in args.tag.split(",")):
+                if not tag:
+                    continue
+                q = q.filter(
+                    GalleryImage.tags.ilike(f"%{tag}%") |
+                    GalleryImage.ai_tags.ilike(f"%{tag}%")
+                )
         q = q.order_by(GalleryImage.created_at.desc()).limit(args.limit)
         emit([_serialize_image(i) for i in q.all()], args)
     finally:
@@ -159,7 +166,7 @@ def _build_parser():
     pl.add_argument("--limit", type=int, default=50)
     pl.add_argument("--favorites", action="store_true")
     pl.add_argument("--album", help="filter by album name")
-    pl.add_argument("--tag", help="substring match against tags column")
+    pl.add_argument("--tag", help="comma-separated tags; each matches manual or AI tags by substring")
     pl.set_defaults(func=cmd_list)
 
     psh = sub.add_parser("show", parents=[common])

scripts/odysseus-mail

@@ -43,7 +43,7 @@ import logging
 import os
 import re
 import sys
-from datetime import datetime
+from datetime import datetime, timezone
 from pathlib import Path
 
 # Anchor the path so the script works when invoked from anywhere.
@@ -136,7 +136,7 @@ def cmd_list(args) -> None:
             emit([], args)
             return
         all_uids = data[0].split()
-        # Newest first, limit
+        # Keep a bounded arrival batch; order its decoded rows by Date below.
         uids = list(reversed(all_uids))[: args.limit]
         out = []
         for uid in uids:
@@ -165,10 +165,15 @@ def cmd_list(args) -> None:
                 sender_name, sender_addr = parseaddr(sender)
                 date_raw = msg.get("Date", "")
                 try:
-                    iso = parsedate_to_datetime(date_raw).isoformat() if date_raw else ""
+                    parsed_date = parsedate_to_datetime(date_raw) if date_raw else None
+                    iso = parsed_date.isoformat() if parsed_date else ""
+                    if parsed_date is not None and parsed_date.tzinfo is None:
+                        parsed_date = parsed_date.replace(tzinfo=timezone.utc)
+                    date_epoch = parsed_date.timestamp() if parsed_date else float("-inf")
                 except Exception:
                     iso = ""
-                out.append({
+                    date_epoch = float("-inf")
+                out.append((date_epoch, {
                     "uid": uid.decode(),
                     "date": iso,
                     "from_addr": sender_addr,
@@ -176,10 +181,11 @@ def cmd_list(args) -> None:
                     "subject": subject,
                     "is_read": "\\Seen" in flags,
                     "is_answered": "\\Answered" in flags,
-                })
+                }))
             except Exception as e:
                 sys.stderr.write(f"warn: skipping uid {uid!r}: {e}\n")
-        emit(out, args)
+        out.sort(key=lambda item: item[0], reverse=True)
+        emit([row for _, row in out], args)
 
 
 # ─── read ────────────────────────────────────────────────────────────

scripts/sanitize_trace_artifacts.py

@@ -77,13 +77,14 @@ def sanitize_body_text(text: str) -> str:
 
     # 1. Try JSON
     trimmed = text.strip()
-    if (trimmed.startswith("{") and trimmed.endswith("}")) or (trimmed.startswith("[") and trimmed.endswith("]")):
+    if trimmed.startswith(("{", "[")):
         try:
             parsed = json.loads(trimmed)
             sanitized = sanitize_structured_obj(parsed)
             return json.dumps(sanitized, ensure_ascii=False)
-        except Exception:
+        except Exception as exc:
-            pass
+            if SENSITIVE_KEY_PATTERN.search(trimmed) or any(p in trimmed.lower() for p in SENSITIVE_PARAM_NAMES):
+                raise ValueError(f"Malformed sensitive JSON request body: {exc}") from exc
 
     # 2. Try URL-encoded form data
     if "=" in trimmed and not trimmed.startswith("<"):
@@ -98,8 +99,13 @@ def sanitize_body_text(text: str) -> str:
                     else:
                         out_pairs.append((k, v))
                 return urllib.parse.urlencode(out_pairs)
-        except Exception:
+        except Exception as exc:
-            pass
+            if SENSITIVE_KEY_PATTERN.search(trimmed):
+                raise ValueError(f"Malformed sensitive form request body: {exc}") from exc
+
+    # If unparsed body text contains sensitive keys or tokens, fail closed
+    if SENSITIVE_KEY_PATTERN.search(trimmed) or any(p in trimmed.lower() for p in SENSITIVE_PARAM_NAMES):
+        raise ValueError("Malformed or unsupported sensitive request body in trace entry")
 
     return text
 
@@ -207,7 +213,11 @@ def sanitize_structured_obj(obj: Any) -> Any:
 
 
 def sanitize_jsonl_content(data: bytes) -> bytes:
-    """Sanitize newline-delimited JSON Lines (.trace, .network, .stacks)."""
+    """Sanitize newline-delimited JSON Lines (.trace, .network, .stacks).
+    
+    If any non-empty record cannot be decoded as valid UTF-8 or parsed as valid JSON,
+    or is malformed, raise ValueError fail-closed so the affected archive is permanently removed.
+    """
     lines = data.split(b"\n")
     out_lines = []
     for line in lines:
@@ -216,10 +226,8 @@ def sanitize_jsonl_content(data: bytes) -> bytes:
             continue
         try:
             parsed = json.loads(line.decode("utf-8"))
-        except (json.JSONDecodeError, UnicodeDecodeError):
+        except (json.JSONDecodeError, UnicodeDecodeError) as err:
-            # Not valid JSON, keep as-is
+            raise ValueError(f"Malformed or unsupported JSONL trace record: {err}")
-            out_lines.append(line)
-            continue
         sanitized = sanitize_structured_obj(parsed)
         out_lines.append(json.dumps(sanitized, ensure_ascii=False).encode("utf-8"))
     return b"\n".join(out_lines)
@@ -275,11 +283,10 @@ def sanitize_trace_archive(zp: str) -> bool:
                     elif filename_lower.endswith((".json", ".har")):
                         try:
                             parsed = json.loads(content.decode("utf-8"))
-                            sanitized = sanitize_structured_obj(parsed)
+                        except (json.JSONDecodeError, UnicodeDecodeError) as exc:
-                            zout.writestr(item, json.dumps(sanitized, ensure_ascii=False).encode("utf-8"))
+                            raise ValueError(f"Malformed JSON in archive member {item.filename}: {exc}") from exc
-                        except Exception:
+                        sanitized = sanitize_structured_obj(parsed)
-                            # If invalid JSON, treat as text line-by-line
+                        zout.writestr(item, json.dumps(sanitized, ensure_ascii=False).encode("utf-8"))
-                            zout.writestr(item, sanitize_jsonl_content(content))
                     else:
                         # Unknown binary or text member: exclude to fail closed
                         continue

scripts/verify_notes_browser_runtime.py

@@ -11,6 +11,38 @@ import time
 from urllib.request import urlopen
 
 ROOT = Path(__file__).resolve().parents[1]
+if str(ROOT) not in sys.path:
+    sys.path.insert(0, str(ROOT))
+
+
+def _stop_owned_server(server, identity):
+    """Close only identity-bound descendants, including MCP's new sessions.
+
+    Capture before signalling the parent: an orphaned stderr holder can no
+    longer be attributed by ancestry. Never search by executable name or kill
+    the caller's process group.
+    """
+    from src import process_lifecycle as lifecycle
+
+    descendants = (lifecycle.bind_descendants([identity.pid])
+                   if identity.verdict() == lifecycle.OWNED else [])
+    if os.name == "nt":
+        if identity.verdict() == lifecycle.OWNED:
+            lifecycle.taskkill_tree(identity.pid)
+    else:
+        outcome = lifecycle.terminate_identities(
+            [*(entry.identity for entry in reversed(descendants)), identity],
+            steps=lifecycle.term_kill_steps(10),
+        )
+        if outcome.survivors or outcome.unverified:
+            raise RuntimeError(f"Disposable Notes cleanup could not verify exit: {outcome}")
+    try:
+        logs, _ = server.communicate(timeout=2)
+    except subprocess.TimeoutExpired as error:
+        # Do not turn an unidentified pipe holder into an unbounded wait.
+        server.stdout.close()
+        raise RuntimeError("Disposable Notes output pipe remained open after owned cleanup") from error
+    return logs
 
 
 def main():
@@ -36,7 +68,10 @@ def main():
         server = subprocess.Popen(
             [sys.executable, "-m", "uvicorn", "app:app", "--host", "127.0.0.1", "--port", str(port)],
             cwd=target, env=env, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True,
+            start_new_session=(os.name != "nt"),
         )
+        from src.process_lifecycle import ProcessIdentity
+        server_identity = ProcessIdentity.capture(server.pid)
         result = None
         try:
             origin = f"http://127.0.0.1:{port}"
@@ -62,12 +97,7 @@ def main():
                 print(result.stderr, file=sys.stderr)
             return result.returncode
         finally:
-            server.terminate()
+            logs = _stop_owned_server(server, server_identity)
-            try:
-                logs, _ = server.communicate(timeout=10)
-            except subprocess.TimeoutExpired:
-                server.kill()
-                logs, _ = server.communicate()
             if result is None or result.returncode:
                 print(logs[-8000:], file=sys.stderr)
 

services/hwfit/hf_discovery.py

@@ -202,17 +202,35 @@ def _infer_use_case(repo_id, pipeline_tag):
     return "general"
 
 
+# HF collections mix in models no Cookbook LLM backend can serve (vision
+# encoders, time-series forecasters, image/video generators). Image models
+# have their own registry, so keep these out of the LLM catalog.
+_NON_LLM_PIPELINE_TAGS = frozenset({
+    "time-series-forecasting",
+    "zero-shot-image-classification",
+    "image-classification",
+    "image-feature-extraction",
+    "text-to-image",
+    "image-to-video",
+    "text-to-video",
+})
+
+
 def _entry_from_collection_item(collection, item, source):
     repo_id = item.get("id") or ""
     if item.get("type") != "model" or not repo_id:
         return None
+    if (item.get("pipeline_tag") or "").strip().lower() in _NON_LLM_PIPELINE_TAGS:
+        return None
     repo_prefix = source.get("repo_prefix")
     if repo_prefix and not repo_id.startswith(repo_prefix):
         return None
     raw_params = item.get("numParameters") or 0
-    active = None
+    # Read the name even when HF gives a total: it is the only source of the
+    # MoE active size (e.g. "-35B-A3B"), which the speed estimate depends on.
+    name_params, active = _parse_params_from_name(repo_id)
     if not raw_params:
-        raw_params, active = _parse_params_from_name(repo_id)
+        raw_params = name_params
     param_label, raw_params = _format_params(raw_params)
     if not raw_params:
         return None

services/memory/skill_format.py

@@ -84,10 +84,14 @@ _FM_KEY_RE = re.compile(r"^([a-z_][a-z0-9_]*):\s*(.*)$", re.IGNORECASE)
 _FM_BLOCK_LIST_RE = re.compile(r"^\s*-\s*(.*)$")
 
 
-def _parse_scalar(raw: str) -> Any:
+def _parse_scalar(raw: str, *, string_only: bool = False) -> Any:
     raw = raw.strip()
     if raw == "":
         return ""
+    # String fields keep their spelling; quoted values still use the decoder
+    # below so existing escapes round-trip as before.
+    if string_only and not (raw[0] == raw[-1] and raw[0] in ("'", '"')):
+        return raw
     if raw.startswith("[") and raw.endswith("]"):
         inner = raw[1:-1].strip()
         if not inner:
@@ -171,7 +175,7 @@ def parse_frontmatter(text: str) -> tuple[Dict[str, Any], str]:
                 pending_key = key
                 fm[key] = []
             else:
-                fm[key] = _parse_scalar(val)
+                fm[key] = _parse_scalar(val, string_only=(key == "version"))
                 pending_key = None
             continue
         m2 = _FM_BLOCK_LIST_RE.match(line)

src/ai_interaction.py

@@ -1207,7 +1207,7 @@ async def do_generate_image(content: str, session_id: Optional[str] = None, owne
       Line 1: prompt describing the image
       Line 2: model name (optional, default auto-detects: prefers gpt-image-1.5 > gpt-image-1)
       Line 3: size (optional, defaults to 1024x1024)
-      Line 4: quality (optional, defaults to medium — options: low, medium, high, auto)
+      Line 4: quality (optional, defaults to configured quality or medium — low, medium, high, auto)
     """
     import base64
     import httpx
@@ -1225,13 +1225,13 @@ async def do_generate_image(content: str, session_id: Optional[str] = None, owne
         prompt = str(args.get('prompt') or '').strip()
         model_spec = str(args.get('model') or '').strip()
         size = str(args.get('size') or '1024x1024')
-        quality = str(args.get('quality') or 'medium')
+        quality = str(args.get('quality') or '')
     else:
         lines = content.strip().split("\n")
         prompt = lines[0].strip() if lines else ""
         model_spec = lines[1].strip() if len(lines) > 1 and lines[1].strip() else ""
         size = lines[2].strip() if len(lines) > 2 and lines[2].strip() else "1024x1024"
-        quality = lines[3].strip() if len(lines) > 3 and lines[3].strip() else "medium"
+        quality = lines[3].strip() if len(lines) > 3 and lines[3].strip() else ""
 
     if not prompt:
         return {"error": "Image prompt is required (line 1)"}
@@ -1249,8 +1249,8 @@ async def do_generate_image(content: str, session_id: Optional[str] = None, owne
     # Use admin-configured model/quality if not specified by the tool call
     if not model_spec:
         model_spec = _settings.get("image_model", "")
-    if quality == "medium" and _settings.get("image_quality"):
+    if not quality:
-        quality = _settings["image_quality"]
+        quality = _settings.get("image_quality") or "medium"
 
     # Auto-detect best available image model if still not set
     if not model_spec:
@@ -1482,7 +1482,7 @@ async def do_edit_image(
     session_id: Optional[str] = None,
     owner: Optional[str] = None,
     size: str = "auto",
-    quality: str = "medium",
+    quality: Optional[str] = None,
     progress_callback: Optional[Callable[[Dict[str, Any]], Awaitable[None]]] = None,
     incognito: bool = False,
 ) -> Dict:
@@ -1509,8 +1509,8 @@ async def do_edit_image(
 
     if not model_spec:
         model_spec = _settings.get("image_model", "")
-    if quality == "medium" and _settings.get("image_quality"):
+    if quality is None:
-        quality = _settings["image_quality"]
+        quality = _settings.get("image_quality") or "medium"
     if not model_spec:
         return {"error": "No image model selected for image editing"}
 

src/llm_core.py

@@ -432,12 +432,15 @@ class _DegenerateStreamGuard:
     "Summer Summer ..."). This is not a useful response and can burn context,
     browser memory, and GPU time. Keep the guard conservative: only fire on long
     same-token runs or a very dominant repeated token in the recent window.
+    Repetition made only of numbers is data (a zero or identity matrix, a
+    padded buffer), so it only stops at a long all-numeric backstop.
     """
 
     def __init__(self, model: str):
         self.model = model or "model"
         self.last_token = ""
         self.same_run = 0
+        self.numeric_run = 0
         self.recent_tokens: List[str] = []
         self.periodic_tokens: List[str] = []
         self.total_chars = 0
@@ -458,6 +461,7 @@ class _DegenerateStreamGuard:
             else:
                 self.last_token = token
                 self.same_run = 1
+            self.numeric_run = self.numeric_run + 1 if token.isdigit() else 0
             self.recent_tokens.append(token)
             self.periodic_tokens.append(token)
         if len(self.recent_tokens) > 96:
@@ -466,14 +470,20 @@ class _DegenerateStreamGuard:
             self.periodic_tokens = self.periodic_tokens[-2048:]
 
         reason = None
-        if self.same_run >= 28 and self.total_chars >= 100:
+        if self.same_run >= 28 and self.total_chars >= 100 and not self.last_token.isdigit():
             reason = f"repeated '{self.last_token}' {self.same_run} times"
         elif len(self.recent_tokens) >= 72:
             top = max(set(self.recent_tokens), key=self.recent_tokens.count)
             count = self.recent_tokens.count(top)
-            if count >= 60 and count / max(len(self.recent_tokens), 1) >= 0.78:
+            if count >= 60 and count / max(len(self.recent_tokens), 1) >= 0.78 and not top.isdigit():
                 reason = f"repeated '{top}' {count}/{len(self.recent_tokens)} recent tokens"
-        if not reason and len(self.recent_tokens) >= 48:
+        if (
+            not reason
+            and len(self.recent_tokens) >= 48
+            # A window that is mostly numbers is matrix/table data; its
+            # periodicity is real. Mixed loops still hit the exact-block check.
+            and sum(t.isdigit() for t in self.recent_tokens) * 2 < len(self.recent_tokens)
+        ):
             # Detect a periodic suffix instead of counting repeated n-grams.
             # Reused property names are normal in CSS/SVG/JSON; a collapsed
             # generation repeats nearly every token at a short fixed period.
@@ -510,10 +520,13 @@ class _DegenerateStreamGuard:
                 period = current_anchor - previous
                 if period < 24 or len(history) < period * 2:
                     continue
-                if history[-2 * period:-period] == history[-period:]:
+                if history[-2 * period:-period] == history[-period:] and not all(t.isdigit() for t in history[-period:]):
                     reason = f"repeated an exact {period}-token block twice"
                     break
 
+        if not reason and self.numeric_run >= 2048:
+            reason = f"emitted {self.numeric_run} numbers in a row"
+
         if not reason:
             return None
 

src/mcp_manager.py

@@ -10,6 +10,7 @@ import logging
 import os
 import re
 import asyncio 
+import contextlib
 from typing import Any, Dict, List, Optional, Set, Tuple
 from src.database import McpServer, SessionLocal
 
@@ -383,12 +384,21 @@ class McpManager:
 
             provider = build_provider(server_id, url, on_redirect=_on_redirect)
             stack = AsyncExitStack()
-            transport = await stack.enter_async_context(streamablehttp_client(url, auth=provider))
+            try:
-            read_stream, write_stream, _get_session_id = transport
+                transport = await stack.enter_async_context(streamablehttp_client(url, auth=provider))
-            session = await stack.enter_async_context(ClientSession(read_stream, write_stream))
+                read_stream, write_stream, _get_session_id = transport
-            await session.initialize()
+                session = await stack.enter_async_context(ClientSession(read_stream, write_stream))
+                await session.initialize()
 
-            tools_result = await session.list_tools()
+                tools_result = await session.list_tools()
+            except BaseException:
+                # Close the half-open transport here, in the task that entered
+                # it. Left for the garbage collector, its AnyIO cancel scope is
+                # exited from the wrong task and the event loop keeps spinning
+                # (#5518). Its own teardown error would only mask the real one.
+                with contextlib.suppress(Exception):
+                    await stack.aclose()
+                raise
             tools = []
             for tool in tools_result.tools:
                 tools.append({

src/model_discovery.py

@@ -193,7 +193,10 @@ class ModelDiscovery:
             data = r.json()
             # Some OpenAI-compatible servers return a bare list, not {"data": [...]}.
             items = data if isinstance(data, list) else ((data or {}).get("data") or [])
-            ids = [m.get("id") for m in items if isinstance(m, dict) and m.get("id")]
+            ids = [
+                m["id"] for m in items
+                if isinstance(m, dict) and isinstance(m.get("id"), str) and m["id"]
+            ]
             if ids:
                 return {
                     "host": host,

src/pdf_form_doc.py

@@ -102,8 +102,10 @@ _PLAIN_FRONT_MATTER_RE = re.compile(
 #   - **label:** value <!-- field=NAME-ENC type=text -->
 #   - **label** [opts]: value <!-- field=NAME-ENC type=choice -->
 #   - [x] **label** <!-- field=NAME-ENC type=checkbox -->
+# Accept Unicode word characters for older saved markers, whose encoder left
+# non-ASCII letters/numbers literal. Newly rendered names are ASCII-safe.
 _FIELD_BULLET_RE = re.compile(
-    r'^\s*-\s+(?P<body>.*?)\s*<!--\s*field=(?P<name>[A-Za-z0-9_.%-]+)\s+type=(?P<type>\w+)\s*-->\s*$'
+    r'^\s*-\s+(?P<body>.*?)\s*<!--\s*field=(?P<name>[\w.%-]+)\s+type=(?P<type>\w+)\s*-->\s*$'
 )
 
 
@@ -115,7 +117,7 @@ def _encode_name(name: str) -> str:
     """
     out = []
     for ch in name or "":
-        if ch.isalnum() or ch in ("_", ".", "-"):
+        if (ch.isascii() and ch.isalnum()) or ch in ("_", ".", "-"):
             out.append(ch)
         else:
             for b in ch.encode("utf-8"):

src/session_actions.py

@@ -12,16 +12,6 @@ from datetime import datetime, timedelta, timezone
 
 logger = logging.getLogger(__name__)
 
-# Names that indicate a throwaway/test session
-_THROWAWAY_NAMES = {
-    "test", "testing", "asdf", "asd", "hello", "hi", "hey",
-    "yo", "sup", "hola", "hii", "hiii", "heyo",
-    "foo", "bar", "baz", "tmp", "temp", "scratch", "untitled",
-    "new chat", "delete", "remove", "junk", "trash", "xxx",
-    "abc", "qwerty", "blah", "stuff", "whatever", "idk",
-    "ok", "lol", "bruh", "hmm", "hm", "meh",
-}
-_THROWAWAY_MAX_MESSAGES = 4
 _FRESH_EMPTY_SESSION_GRACE = timedelta(minutes=10)
 _FRESH_SESSION_GRACE = _FRESH_EMPTY_SESSION_GRACE
 
@@ -54,14 +44,15 @@ def is_session_recently_active(row, now=None, grace=_FRESH_SESSION_GRACE) -> boo
 
 
 async def run_auto_sort(owner: str, skip_llm: bool = False, delete_throwaway: bool = True) -> str:
-    """Run session cleanup + (optional) AI folder sort for the given owner.
+    """Remove empty/Incognito sessions and optionally organize saved chats.
 
     Args:
         owner: user whose sessions to process
-        skip_llm: when True, do only Phase 1 (delete empty/throwaway sessions);
+        skip_llm: when True, do only Phase 1 (delete empty/Incognito sessions);
             skip Phase 2 (AI folder assignment). Used by the built-in daily
             background sweep so it never burns LLM tokens.
-        delete_throwaway: when False, only empty/incognito sessions are deleted.
+        delete_throwaway: retained for caller compatibility. Ordinary sessions
+            with saved messages are always preserved, regardless of this flag.
 
     Returns a human-readable summary of what was done.
     """
@@ -71,7 +62,7 @@ async def run_auto_sort(owner: str, skip_llm: bool = False, delete_throwaway: bo
 
     db = SessionLocal()
     try:
-        # ── Phase 1: Delete empty/throwaway sessions ──
+        # ── Phase 1: Delete only empty and explicitly Incognito sessions ──
         deleted_empty = 0
         deleted_throwaway = 0
 
@@ -93,47 +84,13 @@ async def run_auto_sort(owner: str, skip_llm: bool = False, delete_throwaway: bo
             if is_session_recently_active(row, now=cleanup_now):
                 continue
 
-            msg_count = db.query(DbMsg.id).filter(
+            has_messages = db.query(DbMsg.id).filter(
                 DbMsg.session_id == row.id
-            ).limit(_THROWAWAY_MAX_MESSAGES + 1).count()
+            ).first() is not None
-            should_delete = False
+            if not has_messages:
-
-            if msg_count == 0:
                 if is_fresh:
                     continue
-                should_delete = True
                 deleted_empty += 1
-            elif delete_throwaway and msg_count <= _THROWAWAY_MAX_MESSAGES:
-                name = (row.name or "").strip().lower()
-                first_msg = db.query(DbMsg.content).filter(
-                    DbMsg.session_id == row.id, DbMsg.role == "user"
-                ).order_by(DbMsg.timestamp).first()
-                first_text = (first_msg[0] or "").strip().lower() if first_msg else ""
-                assistant_count = db.query(DbMsg.id).filter(
-                    DbMsg.session_id == row.id, DbMsg.role == "assistant"
-                ).limit(1).count()
-
-                if name in _THROWAWAY_NAMES or name.startswith("chat:") or first_text in _THROWAWAY_NAMES:
-                    should_delete = True
-                    deleted_throwaway += 1
-                elif msg_count == 1 and assistant_count == 0:
-                    should_delete = True
-                    deleted_throwaway += 1
-                elif msg_count <= 4 and first_text and len(first_text.split()) <= 8 and len(first_text) <= 80:
-                    # Short trivial chats — e.g. "write hi to a friend" → "Hi!"
-                    should_delete = True
-                    deleted_throwaway += 1
-                else:
-                    # Aggressive: total message text under 250 chars combined = trivial
-                    msg_rows = db.query(DbMsg.content).filter(
-                        DbMsg.session_id == row.id
-                    ).all()
-                    total_chars = sum(len(m[0] or "") for m in msg_rows)
-                    if total_chars <= 250:
-                        should_delete = True
-                        deleted_throwaway += 1
-
-            if should_delete:
                 db.delete(row)
 
         if deleted_empty or deleted_throwaway:

src/task_scheduler.py

@@ -132,6 +132,23 @@ async def _cached(key: Tuple, ttl: float, fetch: Callable[[], Awaitable[Any]]) -
             _shared_cache_pending.pop(key, None)
 
 
+def is_valid_cron(cron_expression: str) -> bool:
+    """True for a standard 5-field cron expression that croniter can parse.
+
+    croniter also accepts 6- and 7-field expressions with a seconds field, but
+    the scheduler works at one-minute resolution and would run such a task
+    every second.
+    """
+    if len((cron_expression or "").split()) != 5:
+        return False
+    try:
+        from croniter import croniter
+        croniter(cron_expression)
+    except Exception:
+        return False
+    return True
+
+
 def compute_next_run(schedule: str, scheduled_time: str,
                      scheduled_day: int = None,
                      scheduled_date: datetime = None,
@@ -175,6 +192,9 @@ def compute_next_run(schedule: str, scheduled_time: str,
         return dt.astimezone(timezone.utc).replace(tzinfo=None)
 
     if schedule == "cron" and cron_expression:
+        if not is_valid_cron(cron_expression):
+            logger.warning(f"Invalid cron expression '{cron_expression}'")
+            return None
         try:
             from croniter import croniter
             cron = croniter(cron_expression, now)
@@ -306,7 +326,7 @@ def _digest_windows(now):
 
 
 def _checkin_calendar_events(db, owner, start, end):
-    """Calendar events in [start, end] for ONE owner, for the check-in digest.
+    """Calendar events in [start, end) for ONE owner, for the check-in digest.
 
     Ownership lives on CalendarCal.owner; events inherit it via calendar_id.
     The digest query had no owner scope, so it pulled EVERY user's events into
@@ -320,7 +340,7 @@ def _checkin_calendar_events(db, owner, start, end):
         .filter(
             _CC.owner == owner,
             _CE.dtstart >= start,
-            _CE.dtstart <= end,
+            _CE.dtstart < end,
             _CE.status != "cancelled",
         )
         .order_by(_CE.dtstart)

src/tool_schemas.py

@@ -41,6 +41,7 @@ _REQUIRED_NATIVE_TOOL_ARGS = {
     "write_file": ("path",),
     "edit_file": ("path",),
     "apply_patch": ("patch_text", "patchText", "patch"),
+    "generate_image": ("prompt",),
     "host_shell": ("command",),
 }
 

static/css/01-agent-chat.css

@@ -4592,7 +4592,9 @@ body.bg-pattern-ascii-fireflies {
       width:100%; background:var(--bg); color:var(--fg);
       border:1px solid var(--border); border-radius:6px;
       padding:8px; font-family:inherit; font-size:0.95rem;
-      resize:vertical; outline:none;
+      resize:both; outline:none;
+      field-sizing:content;
+      max-width:100%; box-sizing:border-box;
       min-height:80px;
     }
     .edit-textarea:focus { border-color:var(--red); }

static/css/17-research.css

@@ -2690,11 +2690,14 @@ body.theme-frosted .modal {
     max-height: 95dvh !important;
   }
 
+  /* Notes and Documents must respect the same viewport and safe-area
+     top as other mobile full-height sheets. Keep these authoritative rules
+     consistent with 05-documents.css and 14-email.css. */
   body.doc-view .doc-editor-pane,
   body.notes-view .notes-pane {
-    inset: auto 0 0 !important;
+    inset: var(--sheet-top) 0 0 !important;
-    height: 95dvh !important;
+    height: calc(var(--app-height) - var(--sheet-top)) !important;
-    max-height: 95dvh !important;
+    max-height: calc(var(--app-height) - var(--sheet-top)) !important;
   }
 }
 

static/index.html

@@ -16,13 +16,6 @@
   <link rel="apple-touch-icon" href="/static/icons/icon-192.png">
   <script nonce="{{CSP_NONCE}}">
   window._odysseusLoadTime = Date.now();
-  // Stops iOS focus-zoom on every field; iOS still allows pinch-zoom, other platforms would not.
-  (function(){
-    var ua = navigator.userAgent || '';
-    var isIOS = /iPad|iPhone|iPod/.test(ua) || (/Macintosh/.test(ua) && navigator.maxTouchPoints > 1);
-    var vp = document.querySelector('meta[name="viewport"]');
-    if (isIOS && vp) vp.setAttribute('content', vp.getAttribute('content') + ', maximum-scale=1');
-  })();
   (function(){
     try {
       var t = JSON.parse(localStorage.getItem('odysseus-theme'));

static/js/calendar.js

@@ -771,7 +771,13 @@ function _getModal() {
       <button class="cal-side-nav cal-side-nav-next" id="cal-side-next" title="Next" aria-label="Next">&rsaquo;</button>
     </div>`;
   document.body.appendChild(_modal);
-  _modal.querySelector('#cal-close').addEventListener('click', closeCalendar);
+  _modal.querySelector('#cal-close').addEventListener('click', () => {
+    if (_modal.querySelector('.cal-form')) {
+      _render();
+      return;
+    }
+    closeCalendar();
+  });
   _modal.querySelector('#cal-side-prev')?.addEventListener('click', (e) => {
     e.preventDefault();
     e.stopPropagation();

static/js/compare/sse.js

@@ -0,0 +1,65 @@
+function streamError(payload) {
+  const detail = payload?.data ?? payload;
+  const candidates = [detail?.message, detail?.text, detail?.error?.message,
+    detail?.error, payload?.message, payload?.text, payload?.error?.message,
+    payload?.error, typeof payload === 'string' ? payload : null];
+  const message = candidates.find(value => typeof value === 'string' && value.trim());
+  return new Error(message || 'Model request failed.');
+}
+
+/** Decode complete SSE frames, requiring the backend's terminal DONE event. */
+export async function consumeCompareEvents(response, onEvent, onActivity) {
+  const reader = response.body.getReader();
+  const decoder = new TextDecoder();
+  let line = '', event = 'message', data = [], skipLF = false;
+  try {
+    while (true) {
+      const { done, value } = await reader.read();
+      if (!done) onActivity?.();
+      let chunk = decoder.decode(value, { stream: !done });
+      if (skipLF && chunk.length) {
+        if (chunk.startsWith('\n')) chunk = chunk.slice(1);
+        skipLF = false;
+      }
+      const separators = /\r\n|\r|\n/g;
+      let start = 0, separator;
+      while ((separator = separators.exec(chunk))) {
+        line += chunk.slice(start, separator.index);
+        start = separator.index + separator[0].length;
+        skipLF = separator[0] === '\r' && start === chunk.length;
+        if (!line) {
+          if (data.length || event === 'error') {
+            const raw = data.join('\n');
+            if (event !== 'error' && raw === '[DONE]') return;
+            let payload;
+            try { payload = JSON.parse(raw); }
+            catch (_) {
+              if (event === 'error') throw streamError(raw);
+              throw new Error('Invalid model stream data.');
+            }
+            if (event === 'error' || payload?.type === 'error') throw streamError(payload);
+            if (!payload || typeof payload !== 'object' || Array.isArray(payload)) {
+              throw new Error('Invalid model stream data.');
+            }
+            await onEvent(payload);
+          }
+          event = 'message';
+          data = [];
+        } else if (!line.startsWith(':')) {
+          const colon = line.indexOf(':');
+          const field = colon < 0 ? line : line.slice(0, colon);
+          let value = colon < 0 ? '' : line.slice(colon + 1);
+          if (value.startsWith(' ')) value = value.slice(1);
+          if (field === 'data') data.push(value);
+          else if (field === 'event') event = value || 'message';
+        }
+        line = '';
+      }
+      line += chunk.slice(start);
+      if (done) throw new Error('Stream ended before completion.');
+    }
+  } finally {
+    try { await reader.cancel(); } catch (_) {}
+    reader.releaseLock();
+  }
+}

static/js/compare/stream.js

@@ -1,5 +1,6 @@
 // compare/stream.js — SSE streaming to panes
 import state from './state.js';
+import { consumeCompareEvents } from './sse.js';
 import { addFinishBadge } from './vote.js?v=20260828resendcaldrag1';
 import { getModelCost, renderAskUserCard, safeDisplayImageSrc } from '../chatRenderer.js?v=20260914metricssummary1';
 import markdownModule from '../markdown.js';
@@ -472,25 +473,9 @@ async function streamToPane(paneIdx, sessionId, message, aiMsgEl, opts) {
     });
     if (!response.ok) throw new Error(`HTTP ${response.status}`);
 
-    const reader = response.body.getReader();
+    await consumeCompareEvents(response,
-    const decoder = new TextDecoder();
+      (json) => {
-    let buffer = '';
-
-    while (true) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      _resetIdleTimeout();  // any chunk = stream is alive
-
-      buffer += decoder.decode(value, { stream: true });
-      const lines = buffer.split('\n');
-      buffer = lines.pop() || '';
-
-      for (const line of lines) {
-        if (!line.startsWith('data: ')) continue;
-        const data = line.slice(6);
-        if (data === '[DONE]') break;
         try {
-          const json = JSON.parse(data);
           if (json.type === 'metrics') {
             metrics = json.data;
 
@@ -672,7 +657,7 @@ async function streamToPane(paneIdx, sessionId, message, aiMsgEl, opts) {
             if (hist) hist.scrollTop = hist.scrollHeight;
           } else if (json.delta) {
             // Skip text deltas if we already rendered an image
-            if (aiMsgEl._imageData) continue;
+            if (aiMsgEl._imageData) return;
             // Capture TTFT on very first text delta
             if (!accumulated && !_ttft) _ttft = performance.now() - _timerStart;
             // On first delta, destroy spinner and prepare text area
@@ -701,8 +686,7 @@ async function streamToPane(paneIdx, sessionId, message, aiMsgEl, opts) {
             _scheduleLiveRender(target);
           }
         } catch (e) { console.warn('Compare stream render error:', e); }
-      }
+      }, _resetIdleTimeout);
-    }
 
     streamOk = true;
     // Destroy any remaining spinner
@@ -900,9 +884,15 @@ async function streamToPane(paneIdx, sessionId, message, aiMsgEl, opts) {
       }
     } else {
       console.error('Compare stream error:', error);
-      aiBody.innerHTML = '<span style="color:var(--color-error);">Error: ' + escapeHtml(error.message) + '</span>';
+      awaitingChoice = false;
+      const message = state._blindMode ? 'Model request failed. Try again.' : error.message;
+      aiBody.innerHTML = '<span style="color:var(--color-error);">Error: ' + escapeHtml(message) + '</span>';
     }
   } finally {
+    if (aiMsgEl._spinner) {
+      aiMsgEl._spinner.destroy();
+      aiMsgEl._spinner = null;
+    }
     clearTimeout(timeoutId);
     _timerDone = true;
     cancelAnimationFrame(_rafId);

static/js/cookbook.js

@@ -1195,11 +1195,12 @@ async function _fetchDependencies() {
       // For backends with a recipe catalog (vllm / sglang / llama_cpp),
       // append a caret button that toggles a per-row recipe panel below.
       const hasRecipe = RECIPE_BACKENDS.has(pkg.name);
-      // Standalone recipe-caret button removed — the "Pick install
+      // The caret is the only way to open the recipe panel (the Installed ▾
-      // command" action lives inside the Installed ▾ dropdown menu
+      // menu has no recipe item), and the HW Fit / serve-card deep links
-      // (see _showDepMenu) so each row only has ONE caret to click.
+      // open it by clicking this caret (#6747).
-      // Kept the variable so downstream concat code stays the same.
+      const recipeCaret = hasRecipe
-      const recipeCaret = '';
+        ? `<button class="cookbook-dep-tag cookbook-dep-recipe-caret" data-dep-recipe-toggle="${esc(pkg.name)}" title="Pick a model to see the exact install commands" aria-expanded="false" style="background:none;border:1px solid var(--border);padding:2px 6px;display:inline-flex;align-items:center;cursor:pointer;"><svg width="11" height="11" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" style="transition:transform 0.15s"><polyline points="6 9 12 15 18 9"/></svg></button>`
+        : '';
       const recipePanel = hasRecipe ? _recipePanelHtml(pkg.name) : '';
       // When llama_cpp (or any future engine) reports build_deps_missing
       // from its system_prereqs probe, surface a one-tap install button

static/js/document.js

@@ -5232,7 +5232,7 @@ import { attachColorPicker } from './colorPicker.js?v=20260910eyedropper1';
   }
 
   async function _handleAttachUpload(e) {
-    const files = e.target.files;
+    const files = Array.from(e.target.files || []);
     e.target.value = ''; // reset for next upload
     await _uploadComposeFiles(files);
   }
@@ -5963,7 +5963,7 @@ import { attachColorPicker } from './colorPicker.js?v=20260910eyedropper1';
   }
 
   async function _handleMarkdownImageUpload(e) {
-    const files = e.target.files;
+    const files = Array.from(e.target.files || []);
     e.target.value = '';
     await _uploadMarkdownImages(files);
   }

static/js/emailLibrary/index.js

@@ -2650,7 +2650,14 @@ export function openEmailLibrary(opts = {}) {
   }
 
   // Wire events
-  document.getElementById('email-lib-close').addEventListener('click', closeEmailLibrary);
+  document.getElementById('email-lib-close').addEventListener('click', () => {
+    const modal = document.getElementById('email-lib-modal');
+    if (modal?.classList.contains('email-reading') || modal?.querySelector('.doclib-card-expanded')) {
+      _exitEmailReaderModeForList();
+      return;
+    }
+    closeEmailLibrary();
+  });
   document.getElementById('email-settings-header-back')?.addEventListener('click', _hideEmailSettingsPage);
 
   // Clicking the modal header (anywhere except buttons/inputs) collapses

static/js/emailLibrary/utils.js

@@ -73,16 +73,24 @@ function _isDangerousSrcset(value) {
 
 // Escape + linkify URLs and email addresses. Returns innerHTML-safe markup.
 export function _escLinkify(text) {
-  const escaped = _esc(text);
+  const raw = String(text || '');
-  // URLs: http(s)://... or www....
+  // Match raw text once so generated anchors are never linkified again.
-  const urlRe = /\b((?:https?:\/\/|www\.)[^\s<>"']+[^\s<>"'.,;:!?)\]])/g;
+  const tokenRe = /\b((?:https?:\/\/|www\.)[^\s<>"']+[^\s<>"'.,;:!?)\]])|\b([A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,})\b/g;
-  const mailRe = /\b([A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,})\b/g;
+  const parts = [];
-  return escaped
+  let offset = 0;
-    .replace(urlRe, (m) => {
+  for (const match of raw.matchAll(tokenRe)) {
-      const href = m.startsWith('www.') ? `https://${m}` : m;
+    parts.push(_esc(raw.slice(offset, match.index)));
-      return `<a href="${_attrEsc(href)}" target="_blank" rel="noopener noreferrer">${m}</a>`;
+    const token = match[0];
-    })
+    const isUrl = !!match[1];
-    .replace(mailRe, (m) => `<a href="${_attrEsc(`mailto:${m}`)}">${m}</a>`);
+    const href = isUrl
+      ? (token.startsWith('www.') ? `https://${token}` : token)
+      : `mailto:${token}`;
+    const attrs = isUrl ? ' target="_blank" rel="noopener noreferrer"' : '';
+    parts.push(`<a href="${_attrEsc(_esc(href))}"${attrs}>${_esc(token)}</a>`);
+    offset = match.index + token.length;
+  }
+  parts.push(_esc(raw.slice(offset)));
+  return parts.join('');
 }
 
 // Pull display name out of "Name <email@x>"; fallback to local-part of

static/js/galleryEditor.js

@@ -5759,7 +5759,20 @@ function _updateBrushCursor(e) {
     state.cursorEl.style.height = diameterY + 'px';
   }
   state.cursorEl.classList.toggle('ge-brush-cursor--pixel', pixelBrush);
-  state.cursorEl.style.transform = `translate3d(${clientX - diameterX / 2}px, ${clientY - diameterY / 2}px, 0)`;
+  // Updating an inline style on every move invalidates document-wide :has()
+  // selectors. A paused animation keeps the same transform in the rendering
+  // pipeline without mutating the cursor's style attribute on every frame.
+  const cursorTransform = `translate3d(${clientX - diameterX / 2}px, ${clientY - diameterY / 2}px, 0)`;
+  if (!state.cursorEl._positionAnimation || state.cursorEl._positionAnimation.playState === 'idle') {
+    state.cursorEl._positionAnimation = state.cursorEl.animate(
+      [{ transform: cursorTransform }, { transform: cursorTransform }],
+      { duration: 1, fill: 'both' },
+    );
+    state.cursorEl._positionAnimation.pause();
+    state.cursorEl._positionAnimation.currentTime = 0;
+  } else {
+    state.cursorEl._positionAnimation.effect.setKeyframes([{ transform: cursorTransform }, { transform: cursorTransform }]);
+  }
   if (state.cursorEl.style.display !== 'block') {
     state.cursorEl.style.display = 'block';
   }
@@ -14061,7 +14074,20 @@ export function closeEditor(options = {}) {
     const h = state.editorDocClickHandlers.pop();
     try { document.removeEventListener('click', h); } catch {}
   }
-  if (state.cursorEl) { state.cursorEl.remove(); state.cursorEl = null; }
+  if (_cursorRaf) { cancelAnimationFrame(_cursorRaf); _cursorRaf = null; }
+  _pendingCursorEvent = null;
+  _lastCursorRect = null;
+  state._lastCursorDiameterX = null;
+  state._lastCursorDiameterY = null;
+  state._lastCursorBorderColor = null;
+  if (state.cursorEl) {
+    if (state.cursorEl._positionAnimation) {
+      try { state.cursorEl._positionAnimation.cancel(); } catch {}
+      state.cursorEl._positionAnimation = null;
+    }
+    state.cursorEl.remove();
+    state.cursorEl = null;
+  }
   try { document.getElementById('ge-cursor-layer')?.remove(); } catch {}
   // Tear down all floating popups + the dock so closing the editor
   // doesn't leave stale chips/panels behind on top of the gallery.

static/js/group.js

@@ -801,7 +801,8 @@ async function _sendRoundRobin(msg, box) {
             headers: { 'Content-Type': 'application/json' },
             body: JSON.stringify({ messages: [{
               role: 'user',
-              content: `[${m._groupName || m.display}]: ${response}`
+              content: `[${m._groupName || m.display}]: ${response}`,
+              metadata: { group_peer: true, group_model: m._groupName || m.display, model: m.mid }
             }]}),
           });
         } catch (e) { console.warn('[group] sync failed:', e); }
@@ -828,7 +829,8 @@ async function _syncAllResponses(holders) {
           headers: { 'Content-Type': 'application/json' },
           body: JSON.stringify({ messages: [{
             role: 'user',
-            content: `[${model._groupName || model.display}]: ${response}`
+            content: `[${model._groupName || model.display}]: ${response}`,
+            metadata: { group_peer: true, group_model: model._groupName || model.display, model: model.mid }
           }]}),
         });
       } catch (e) { /* silent */ }

static/js/init.js

@@ -379,7 +379,6 @@ document.addEventListener('DOMContentLoaded', markComposerUserEdited, { once: tr
       root.classList.toggle('keyboard-open', keyboardOpen);
       if (keyboardOpen) root.style.setProperty('--keyboard-viewport-height', vv.height + 'px');
       else root.style.removeProperty('--keyboard-viewport-height');
-      if (window.scrollY) window.scrollTo(0, 0);
     };
     // iOS fires visualViewport resize only once, mid-animation, and no focusin when
     // the tapped field already had focus (e.g. the composer after a new chat)
@@ -392,7 +391,6 @@ document.addEventListener('DOMContentLoaded', markComposerUserEdited, { once: tr
     window.visualViewport.addEventListener('scroll', _syncKeyboardViewport);
     document.addEventListener('focusin', _syncThroughKeyboardAnimation);
     document.addEventListener('focusout', _syncThroughKeyboardAnimation);
-    window.addEventListener('scroll', () => { if (window.scrollY) window.scrollTo(0, 0); }, { passive: true });
 
     let _lastVVHeight = window.visualViewport.height;
     window.visualViewport.addEventListener('resize', function() {

static/js/memory.js

@@ -163,11 +163,13 @@ function buildCategoryChips() {
   const container = document.getElementById('memory-category-filters');
   if (!container) return;
 
+  const cats = new Set(memories.map(m => m.category || 'fact'));
+  if (activeCategory !== 'all' && !cats.has(activeCategory)) activeCategory = 'all';
+
   // Hide the chip row entirely when there are no memories — no point showing
   // an "all" chip with nothing to filter.
   if (!memories.length) { container.innerHTML = ''; return; }
 
-  const cats = new Set(memories.map(m => m.category || 'fact'));
   const sorted = ['all', ...Array.from(cats).sort()];
   const counts = memories.reduce((result, memory) => {
     const category = memory.category || 'fact';
@@ -277,6 +279,7 @@ export function openMemoryModal(tabName = 'browse') {
       else if (m.default?.loadSkills) m.default.loadSkills(true);
     });
   }
+  if (mode === 'memory') return loadMemories();
 }
 
 function _decodeMemoryAnchorId(memoryId) {
@@ -297,9 +300,8 @@ function _resetMemoryDeepLinkFilters() {
 // Memory search output may expose a shortened id, so prefix matching is
 // intentional here.
 export async function openMemory(memoryId, attempt = 0) {
-  openMemoryModal('browse');
   _resetMemoryDeepLinkFilters();
-  await loadMemories();
+  await openMemoryModal('browse');
   if (renderMemoryList) renderMemoryList();
   const wanted = _decodeMemoryAnchorId(memoryId);
   if (!wanted) return;

static/js/sessions.js

@@ -299,7 +299,18 @@ function _renderHistoryMessage(msg, modelName) {
   } else {
     displayContent = '';
   }
-  if (msg.role === 'user') {
+  let displayRole = msg.role;
+  // Peer turns remain user-role context for the model, but are agent replies
+  // in history. Only explicit identity plus its exact prefix can opt in.
+  if (displayRole === 'user' && meta?.group_peer === true
+      && typeof meta.group_model === 'string' && meta.group_model) {
+    const prefix = `[${meta.group_model}]: `;
+    if (displayContent.startsWith(prefix)) {
+      displayRole = 'assistant';
+      displayContent = displayContent.slice(prefix.length);
+    }
+  }
+  if (displayRole === 'user') {
     displayContent = _stripUserVisionBlocks(displayContent);
     const trimmed = displayContent.trim();
     if (
@@ -316,7 +327,7 @@ function _renderHistoryMessage(msg, modelName) {
       displayContent = `[Doc edit: ${docEditMatch[1]}] ${docEditMatch[3]}`;
     }
   }
-  return _addHistoryMessageWithFullRenderer(msg.role, displayContent, modelName, meta);
+  return _addHistoryMessageWithFullRenderer(displayRole, displayContent, modelName, meta);
 }
 
 function _clearHistoryPager() {

static/js/skills.js

@@ -1260,6 +1260,9 @@ async function _expandSkillCard(card, name) {
       card._md = md || '';
     } else {
       pre.textContent = 'Loading…';
+      const editBtn = [...card.querySelectorAll('.skill-card-preview button')]
+        .find(button => button.textContent.trim() === 'Edit');
+      if (editBtn) editBtn.disabled = true;
       try {
         const md = await _fetchSkillMarkdown(name);
         pre.textContent = md || '(empty)';
@@ -1267,6 +1270,8 @@ async function _expandSkillCard(card, name) {
         card._md = md;
       } catch (e) {
         pre.textContent = `Failed to load SKILL.md: ${e.message || 'Unknown error'}`;
+      } finally {
+        if (editBtn) editBtn.disabled = !card._mdLoaded;
       }
     }
   }
@@ -1275,6 +1280,9 @@ async function _expandSkillCard(card, name) {
 // Swap the read-only <pre> for an editable <textarea> (and back). The
 // Edit button toggles; a Save button commits via the markdown endpoint.
 function _toggleSkillEdit(card, name) {
+  // A loading/error preview is not source text. Menu entry points must obey
+  // the same readiness gate as the disabled footer control.
+  if (!card._mdLoaded) return;
   const preview = card.querySelector('.skill-card-preview');
   if (!preview) return;
   const existing = preview.querySelector('.skill-md-editor');
@@ -1288,6 +1296,9 @@ function _toggleSkillEdit(card, name) {
   ta.className = 'skill-md-editor';
   ta.spellcheck = false;
   ta.value = (card._md != null ? card._md : (pre ? pre.textContent : '')) || '';
+  // Let the source set the intrinsic height; the card's existing flex/max
+  // height keeps long files bounded and the Save footer reachable.
+  ta.rows = Math.max(ta.rows, ta.value.split(/\r\n|\r|\n/).length);
   ta.addEventListener('click', (e) => e.stopPropagation());
   if (pre) pre.style.display = 'none';
   preview.insertBefore(ta, preview.querySelector('.doclib-card-expanded-actions'));

static/js/windowResize.js

@@ -267,8 +267,9 @@ export function makeWindowResizable(content, options = {}) {
           content.style.maxWidth = 'none';
           content.style.maxHeight = 'none';
           const left = parseFloat(content.style.left);
-          if (Number.isFinite(left) && left + w > window.innerWidth - padRight) {
+          if (Number.isFinite(left)) {
-            content.style.left = Math.max(padLeft, window.innerWidth - padRight - w) + 'px';
+            const rightmostLeft = Math.max(padLeft, window.innerWidth - padRight - w);
+            content.style.left = Math.max(padLeft, Math.min(left, rightmostLeft)) + 'px';
           }
         }
       } catch (_) {}

static/sw.js

@@ -7,7 +7,7 @@
 //   - Other static assets (images/fonts/libs): cache-first with bg refresh.
 //   - API / non-GET: never cached.
 // Bump CACHE_NAME whenever the precache list or SW logic changes.
-const CACHE_NAME = 'odysseus-v1114-pr40-editor-precache';
+const CACHE_NAME = 'odysseus-v1115-pr40-editor-precache';
 
 // KaTeX resolves these from its own stylesheet, so caching the CSS without them
 // gives offline math fallback glyphs instead of proper typesetting.
@@ -38,6 +38,7 @@ const KATEX_FONTS = [
 // Both are fetched at install time, in the background. Entries must match the
 // exact URL the browser requests, query string included.
 const PRECACHE = [
+  '/static/js/compare/sse.js',
   '/',
   '/static/css/00-tokens.css?v=20260929csssplit',
   '/static/css/01-agent-chat.css?v=20260929csssplit',

tests/TESTING_STANDARD.md

@@ -99,8 +99,10 @@ cleanup. Specifically:
   `os.environ[...] = ...` that outlives the test.
 - **Current working directory** - never `chdir` without restoring; never assert
   against cwd-relative paths like `./data`. Use a temp workspace helper instead.
-- **Database** - the root `conftest.py` defaults `DATABASE_URL` to an in-memory
+- **Database** - the root `conftest.py` defaults `DATABASE_URL` to a private,
-  SQLite for collection safety. A test that needs a real file-backed DB must opt
+  worker-owned SQLite file for collection safety and consistent schema across
+  TestClient threads. This fallback is not a per-test data fixture. A test that
+  writes or depends on particular rows must opt
   in explicitly via `tests.helpers.sqlite_db.make_temp_sqlite` and bind its
   `SessionLocal` onto the module under test. Do not rely on a persistent
   on-disk DB existing.

tests/cli/test_gallery_cli_tag_filters.py

@@ -0,0 +1,67 @@
+"""Gallery CLI tag filters match both sources and stack like the web library."""
+from datetime import datetime, timedelta
+
+import pytest
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+
+from core.database import Base, GalleryAlbum, GalleryImage
+from tests.helpers.cli_loader import load_script
+
+
+@pytest.fixture
+def gallery_cli(monkeypatch, tmp_path):
+    engine = create_engine(f"sqlite:///{tmp_path / 'gallery.db'}")
+    Base.metadata.create_all(engine)
+    sessions = sessionmaker(bind=engine)
+    with sessions() as db:
+        db.add(GalleryAlbum(id='sea', name='Sea'))
+        for index, (uid, tags, ai_tags, favorite, album) in enumerate([
+            ('manual', 'sunset,beach', '', True, 'sea'),
+            ('ai', '', 'Beach,sunset', False, 'sea'),
+            ('split', 'sunset', 'beach', True, None),
+            ('one', 'sunset', '', False, None),
+            ('other', 'forest', 'moss', False, None),
+            ('inactive', 'sunset,beach', 'sunset,beach', True, 'sea'),
+        ]):
+            db.add(GalleryImage(id=uid, filename=uid+'.png', tags=tags, ai_tags=ai_tags,
+                                favorite=favorite, album_id=album, is_active=uid!='inactive',
+                                created_at=datetime(2026, 1, 1) + timedelta(minutes=index)))
+        db.commit()
+    cli = load_script('odysseus-gallery')
+    monkeypatch.setattr(cli, 'SessionLocal', sessions)
+    output = []
+    monkeypatch.setattr(cli, 'emit', lambda rows, args: output.extend(rows))
+
+    def run(*argv):
+        output.clear()
+        args = cli._build_parser().parse_args(['list', *argv])
+        args.func(args)
+        return [row['id'] for row in output]
+
+    yield run
+    engine.dispose()
+
+
+@pytest.mark.parametrize('tag,expected', [
+    ('sunset', ['one', 'split', 'ai', 'manual']),
+    ('beach,sunset', ['split', 'ai', 'manual']),
+    ('sunset,beach', ['split', 'ai', 'manual']),
+    (', sunset, , BEACH ,', ['split', 'ai', 'manual']),
+    ('forest', ['other']),
+    ('no-match', []),
+    (' , ', ['other', 'one', 'split', 'ai', 'manual']),
+    ('sun', ['one', 'split', 'ai', 'manual']),
+])
+def test_tag_sources_and_stacking(gallery_cli, tag, expected):
+    assert gallery_cli('--tag', tag) == expected
+
+
+@pytest.mark.parametrize('options,expected', [
+    ([], ['other', 'one', 'split', 'ai', 'manual']),
+    (['--tag', 'beach,sunset', '--favorites'], ['split', 'manual']),
+    (['--tag', 'beach,sunset', '--album', 'Sea'], ['ai', 'manual']),
+    (['--tag', 'beach', '--limit', '1'], ['split']),
+])
+def test_tag_filters_preserve_other_list_options(gallery_cli, options, expected):
+    assert gallery_cli(*options) == expected

tests/cli/test_mail_cli_date_order.py

@@ -0,0 +1,66 @@
+"""The bounded arrival batch is displayed by Date, rather than arrival ID."""
+from contextlib import contextmanager
+from types import SimpleNamespace
+
+import pytest
+from tests.cli.test_mail_cli_read_empty_fetch import _load_mail_cli
+
+
+@pytest.mark.parametrize('dates,expected', [
+    (['Wed, 03 Jun 2026 12:00:00 +0000', 'Mon, 01 Jun 2026 09:00:00 +0000'], ['1', '2']),
+    (['Wed, 03 Jun 2026 12:00:00 +0000', 'Wed, 03 Jun 2026 13:00:00 +0200'], ['1', '2']),
+    (['Wed, 03 Jun 2026 12:00:00', 'Wed, 03 Jun 2026 11:00:00 +0000'], ['1', '2']),
+    (['Wed, 03 Jun 2026 12:00:00 +0000', None, 'not a date'], ['1', '3', '2']),
+    (['Wed, 03 Jun 2026 12:00:00 +0000', 'Wed, 03 Jun 2026 14:00:00 +0200'], ['2', '1']),
+    (['Wed, 03 Jun 2026 12:00:00 +0000', 'Thu, 04 Jun 2026 12:00:00 +0000'], ['2', '1']),
+    (['Wed, 31 Dec 1969 12:00:00 +0000', None], ['1', '2']),
+])
+def test_list_sorts_parsed_dates_without_changing_row_schema(monkeypatch, dates, expected):
+    cli = _load_mail_cli(monkeypatch)
+    fetched = []
+
+    class Connection:
+        def select(self, folder, readonly=True):
+            assert readonly is True
+            return 'OK', [str(len(dates)).encode()]
+
+        def search(self, *_):
+            return 'OK', [' '.join(str(i+1) for i in range(len(dates))).encode()]
+
+        def fetch(self, uid, _):
+            fetched.append(uid.decode())
+            date = dates[int(uid)-1]
+            raw = 'From: Sender <sender@example.com>\r\nSubject: Item '+uid.decode()+'\r\n'
+            if date is not None:
+                raw += 'Date: '+date+'\r\n'
+            return 'OK', [(b'1 (FLAGS (\\Seen) RFC822.HEADER)', (raw+'\r\n').encode())]
+
+    @contextmanager
+    def imap(_):
+        yield Connection()
+
+    rows = []
+    monkeypatch.setattr(cli, '_imap', imap)
+    monkeypatch.setattr(cli, 'emit', lambda value, args: rows.extend(value))
+    cli.cmd_list(SimpleNamespace(account='fixture', folder='INBOX', limit=20))
+    assert [row['uid'] for row in rows] == expected
+    assert fetched == [str(i) for i in range(len(dates), 0, -1)]
+    assert all(set(row) == {'uid', 'date', 'from_addr', 'from_name', 'subject', 'is_read', 'is_answered'} for row in rows)
+    assert all(row['is_read'] and not row['is_answered'] for row in rows)
+
+
+def test_list_still_limits_the_arrival_batch(monkeypatch):
+    cli = _load_mail_cli(monkeypatch)
+    fetched = []
+    class Connection:
+        def select(self, *_args, **_kwargs):return 'OK', []
+        def search(self, *_):return 'OK', [b'1 2 3']
+        def fetch(self, uid, _):
+            fetched.append(uid)
+            return 'OK', [(b'3 (FLAGS () RFC822.HEADER)', b'From: x@example.com\r\nSubject: bounded\r\n\r\n')]
+    @contextmanager
+    def imap(_):yield Connection()
+    monkeypatch.setattr(cli, '_imap', imap)
+    monkeypatch.setattr(cli, 'emit', lambda *_:None)
+    cli.cmd_list(SimpleNamespace(account='fixture', folder='INBOX', limit=1))
+    assert fetched == [b'3']

tests/conftest.py

@@ -12,9 +12,17 @@ sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 # File-backed databases remain fixture-owned. Cleanup restores the caller.
 # Keep the existing generated environment-reference locations stable.
 _database_environment = pytest.MonkeyPatch()
-_database_environment.setenv("DATABASE_URL", "sqlite:///:memory:")
 from tests.helpers.worker_runtime import bootstrap_runtime, configure_runtime
 _runtime_environment = bootstrap_runtime()
+# A file keeps the collection schema consistent across TestClient threads;
+# SingletonThreadPool can evict the populated in-memory main-thread database.
+# One patcher owns this variable, including restoration of an inherited URL.
+_database_environment.setenv(
+    "DATABASE_URL",
+    f"sqlite:///{_runtime_environment.root / 'collection.db'}"
+    if _runtime_environment is not None else "sqlite:///:memory:",
+)
+_collection_database_engine = None
 
 # Pre-import real heavy modules BEFORE any test file's module-level stubs can
 # replace them with MagicMock. Some test files (e.g. test_llm_core_sanitize_*)
@@ -26,6 +34,7 @@ try:
     import sqlalchemy.orm  # noqa: F401
     import core.database  # noqa: F401
     import src.database
+    _collection_database_engine = core.database.engine
 except ImportError:
     pass  # not installed - the stubs below will handle it
 
@@ -100,6 +109,10 @@ def pytest_configure(config):
     only registers marker names; it imports no production module.
     """
     config.add_cleanup(_database_environment.undo)
+    if _collection_database_engine is not None:
+        # Registered after runtime cleanup, so connections close before the
+        # owned file is removed (also required on Windows).
+        config.add_cleanup(_collection_database_engine.dispose)
 
     import pathlib
     from tests._taxonomy import discover_markers

tests/e2e/README.md

@@ -1,5 +1,7 @@
 # Browser End-to-End Tests
 
+## Photo Editor Release Gate
+
 The photo editor release gate launches an isolated Odysseus server on port
 `7013` with a temporary SQLite database and disposable authenticated account.
 
@@ -18,3 +20,53 @@ The disposable account is created in the temporary data directory and exercises
 owner-scoped draft and notification routes without using an existing account.
 `PHOTO_EDITOR_E2E_AUTH=0` selects the legacy unauthenticated configuration for
 investigating it; offline-upgrade coverage requires the authenticated mode.
+
+## Mobile and iOS Layout Suite
+
+The dedicated mobile and iOS layout suite exercises viewport behavior,
+accessibility zoom, scrolling, mobile safe-area geometry (Notes and Documents
+panes), and floating window restoration against an isolated local Odysseus server
+running on port `17739`.
+
+```bash
+npm install
+npm run test:ios-layout:chromium
+```
+
+### Profiles and Browser Prerequisites
+
+- **Chromium profiles** (`desktop-chromium`, `phone-chromium-portrait`, `phone-chromium-landscape`, `tablet-chromium`):
+  Provide automated baseline coverage in CI and local development without requiring
+  WebKit dependencies.
+- **WebKit profiles** (`phone-webkit-portrait`, `phone-webkit-landscape`, `tablet-webkit`):
+  Used for platform-oriented WebKit rendering validation. WebKit requires browser binaries:
+
+```bash
+npm run test:ios-layout:install
+# or: npx playwright install webkit
+npm run test:ios-layout:webkit
+```
+
+To run all installed profiles:
+
+```bash
+npm run test:ios-layout
+```
+
+### Evidence and Cleanup
+
+- **Evidence artifacts**: Screenshots are written to `test-results/ios-layout/` by
+  default and traces are written to `test-results/ios-layout-traces/`. Both paths
+  are git-ignored. Set `ODYSSEUS_LAYOUT_EVIDENCE_DIR` to override the screenshot
+  output path.
+- **Teardown**: Disposable application state and database files created in the
+  OS temporary directory are automatically cleaned up on both normal and failure
+  teardown, preserving diagnostic screenshots and failure traces.
+
+### Platform Considerations
+
+Emulated Chromium and WebKit viewports validate deterministic CSS cascade,
+computed layout geometry, viewport meta tags, and JavaScript restoration. However,
+browser emulation is not a physical device substitute: physical iOS Safari and
+standalone installed PWAs should still be verified separately for native gesture
+pinch-zoom, software keyboard dismissal transitions, and dynamic hardware notch/island insets.

tests/e2e/artifact-policy.js

@@ -0,0 +1,19 @@
+const fs = require('node:fs');
+const os = require('node:os');
+const path = require('node:path');
+const marker = '.disposable-ci-fixture.json';
+
+function disposableCIArtifacts() {
+  if (!(process.env.CI || process.env.GITHUB_ACTIONS) || process.env.ODYSSEUS_E2E_DISPOSABLE_ARTIFACTS !== '1') return false;
+  const directory = process.env.PHOTO_EDITOR_E2E_DATA_DIR;
+  if (!directory || process.env.PHOTO_EDITOR_E2E_AUTH === '0') return false;
+  const resolved = path.resolve(directory);
+  if (!path.basename(resolved).startsWith('odysseus-photo-editor-e2e-')) return false;
+  if (![os.tmpdir(), '/tmp', '/var/tmp'].some(base => resolved.startsWith(path.resolve(base) + path.sep))) return false;
+  if (process.env.PHOTO_EDITOR_E2E_DB_PATH !== path.join(resolved, 'app.db')) return false;
+  try {
+    const fixture = JSON.parse(fs.readFileSync(path.join(resolved, marker), 'utf8'));
+    return fixture.synthetic === true && fixture.authenticated === true && fixture.database === path.join(resolved, 'app.db');
+  } catch { return false; }
+}
+module.exports = { disposableCIArtifacts, marker };

tests/e2e/ios-layout-teardown.cjs

@@ -0,0 +1,28 @@
+const fs = require('node:fs');
+const os = require('node:os');
+const path = require('node:path');
+
+/**
+ * Global teardown for the iOS/mobile layout Playwright test suite.
+ * Cleans up only disposable temporary application data directories,
+ * preserving diagnostic artifacts (screenshots and failure traces) in test-results/.
+ */
+module.exports = async () => {
+  const dataDirectory = process.env.ODYSSEUS_LAYOUT_DATA_DIR;
+  if (!dataDirectory || typeof dataDirectory !== 'string') return;
+
+  try {
+    const resolvedData = path.resolve(dataDirectory);
+    const allowedTmpBases = [os.tmpdir(), '/var/tmp', '/tmp'].map(b => path.resolve(b));
+    const isTmp = allowedTmpBases.some(base => resolvedData.startsWith(base + path.sep));
+    const baseName = path.basename(resolvedData);
+
+    if (isTmp && (baseName.startsWith('odysseus-ios-layout-') || baseName.startsWith('odysseus-pr6396-'))) {
+      if (fs.existsSync(resolvedData)) {
+        fs.rmSync(resolvedData, { recursive: true, force: true });
+      }
+    }
+  } catch (err) {
+    console.warn(`[ios-layout-teardown] Warning: failed to clean temporary runtime dir: ${err.message}`);
+  }
+};

tests/e2e/ios-layout.config.cjs

@@ -0,0 +1,148 @@
+// Tests against the running Odysseus app, not a static mock page.
+const fs = require('node:fs');
+const os = require('node:os');
+const path = require('node:path');
+const { defineConfig } = require('@playwright/test');
+
+const repositoryRoot = path.resolve(__dirname, '../..');
+const port = Number(process.env.ODYSSEUS_LAYOUT_PORT || '17739');
+const baseURL = `http://127.0.0.1:${port}`;
+const defaultPython = process.platform === 'win32'
+  ? path.join(repositoryRoot, '.venv', 'Scripts', 'python.exe')
+  : path.join(repositoryRoot, '.venv', 'bin', 'python');
+const python = process.env.ODYSSEUS_TEST_PYTHON
+  || (fs.existsSync(defaultPython) ? defaultPython : 'python3');
+
+const dataDirectory = process.env.ODYSSEUS_LAYOUT_DATA_DIR
+  || fs.mkdtempSync(path.join(os.tmpdir(), 'odysseus-ios-layout-'));
+process.env.ODYSSEUS_LAYOUT_DATA_DIR = dataDirectory;
+
+// Default evidence directory to repository-relative test-results/ios-layout
+// while preserving the environment-variable override if provided.
+const evidenceDir = process.env.ODYSSEUS_LAYOUT_EVIDENCE_DIR
+  || path.join(repositoryRoot, 'test-results', 'ios-layout');
+const outputDir = path.join(repositoryRoot, 'test-results', 'ios-layout-traces');
+
+// Emergency process-exit cleanup for disposable application runtime state
+const cleanDataDirSync = () => {
+  const dir = process.env.ODYSSEUS_LAYOUT_DATA_DIR;
+  if (!dir || typeof dir !== 'string') return;
+  try {
+    const resolved = path.resolve(dir);
+    const allowed = [os.tmpdir(), '/var/tmp', '/tmp'].map(b => path.resolve(b));
+    if (allowed.some(base => resolved.startsWith(base + path.sep))) {
+      const baseName = path.basename(resolved);
+      if (baseName.startsWith('odysseus-ios-layout-') || baseName.startsWith('odysseus-pr6396-')) {
+        if (fs.existsSync(resolved)) {
+          fs.rmSync(resolved, { recursive: true, force: true });
+        }
+      }
+    }
+  } catch (_) {}
+};
+
+process.once('SIGINT', () => { cleanDataDirSync(); process.exit(130); });
+process.once('SIGTERM', () => { cleanDataDirSync(); process.exit(143); });
+
+function isWebKitInstalled() {
+  try {
+    const { webkit } = require('@playwright/test');
+    return fs.existsSync(webkit.executablePath());
+  } catch (_) {
+    return false;
+  }
+}
+
+const chromiumProjects = [
+  {
+    name: 'desktop-chromium',
+    use: { browserName: 'chromium', viewport: { width: 1440, height: 900 } },
+  },
+  {
+    name: 'phone-chromium-portrait',
+    use: { browserName: 'chromium', viewport: { width: 390, height: 844 }, hasTouch: true, isMobile: true },
+  },
+  {
+    name: 'phone-chromium-landscape',
+    use: { browserName: 'chromium', viewport: { width: 844, height: 390 }, hasTouch: true, isMobile: true },
+  },
+  {
+    name: 'tablet-chromium',
+    use: { browserName: 'chromium', viewport: { width: 820, height: 1000 }, hasTouch: true, isMobile: true },
+  },
+];
+
+const webkitProjects = [
+  {
+    name: 'phone-webkit-portrait',
+    use: { browserName: 'webkit', viewport: { width: 390, height: 844 }, hasTouch: true, isMobile: true },
+  },
+  {
+    name: 'phone-webkit-landscape',
+    use: { browserName: 'webkit', viewport: { width: 844, height: 390 }, hasTouch: true, isMobile: true },
+  },
+  {
+    name: 'tablet-webkit',
+    use: { browserName: 'webkit', viewport: { width: 820, height: 1000 }, hasTouch: true, isMobile: true },
+  },
+];
+
+// Determine active projects based on environment and local browser availability
+const requestedBrowser = (process.env.ODYSSEUS_LAYOUT_BROWSER || 'all').toLowerCase();
+let selectedProjects = [];
+
+if (requestedBrowser === 'chromium') {
+  selectedProjects = chromiumProjects;
+} else if (requestedBrowser === 'webkit') {
+  selectedProjects = webkitProjects;
+} else {
+  // 'all': Include chromium always; include webkit if installed or if explicitly targeted via cli args
+  const hasCliWebKitTarget = process.argv.some(arg => arg.toLowerCase().includes('webkit'));
+  if (isWebKitInstalled() || hasCliWebKitTarget) {
+    selectedProjects = [...chromiumProjects, ...webkitProjects];
+  } else {
+    // Only Chromium is available locally (e.g. standard CI / default runner)
+    selectedProjects = chromiumProjects;
+  }
+}
+
+module.exports = defineConfig({
+  testDir: path.join(__dirname, 'ios-layout'),
+  testMatch: ['**/*.spec.cjs'],
+  outputDir,
+  timeout: 90_000,
+  expect: { timeout: 15_000 },
+  retries: 0,
+  workers: 1,
+  reporter: [['line']],
+  globalTeardown: require.resolve('./ios-layout-teardown.cjs'),
+  use: {
+    baseURL,
+    headless: true,
+    serviceWorkers: 'block',
+    screenshot: 'only-on-failure',
+    trace: 'retain-on-failure',
+  },
+  projects: selectedProjects,
+  metadata: { evidenceDir },
+  webServer: {
+    command: `${JSON.stringify(python)} -m uvicorn app:app --host 127.0.0.1 --port ${port}`,
+    cwd: repositoryRoot,
+    url: baseURL,
+    timeout: 120_000,
+    reuseExistingServer: false,
+    env: {
+      ...process.env,
+      AUTH_ENABLED: 'false',
+      // Never connect the disposable app to the host's real ChromaDB.
+      CHROMADB_HOST: '127.0.0.1',
+      CHROMADB_PORT: String(port + 1),
+      LLM_HOST: '127.0.0.1',
+      EMBEDDING_URL: `http://127.0.0.1:${port + 1}/v1/embeddings`,
+      DATABASE_URL: `sqlite:///${path.join(dataDirectory, 'app.db')}`,
+      ODYSSEUS_DATA_DIR: dataDirectory,
+      ODYSSEUS_STARTUP_WARMUPS: '0',
+      RESPONSE_COMPRESSION_ENABLED: 'false',
+    },
+  },
+});

tests/e2e/ios-layout/postmerge-layout.spec.cjs

@@ -0,0 +1,313 @@
+const { test, expect } = require('@playwright/test');
+const fs = require('node:fs');
+const path = require('node:path');
+
+async function openApp(page) {
+  const response = await page.goto('/', { waitUntil: 'domcontentloaded' });
+  expect(response?.ok()).toBeTruthy();
+  // The runtime currently produces two #message nodes. Select the
+  // initialized composer while diagnosing whether both are visible.
+  const composers = await page.locator('textarea#message').evaluateAll(nodes =>
+    nodes.map((el, index) => ({
+      index,
+      initialized: el.dataset.startupPreserveBound === '1',
+      visible: !!el.getClientRects().length &&
+        getComputedStyle(el).visibility !== 'hidden',
+      parent: el.parentElement?.className || '',
+    }))
+  );
+  if (composers.length !== 1) {
+    console.log('Composer DOM diagnostic:', JSON.stringify(composers));
+  }
+  const activeComposer = page.locator(
+    'textarea#message[data-startup-preserve-bound="1"]'
+  );
+  await expect(activeComposer).toHaveCount(1);
+  await expect(activeComposer).toBeVisible();
+  expect(
+    composers.filter(composer => composer.visible).length,
+    'Exactly one composer should be visible'
+  ).toBe(1);
+}
+
+function isMobileViewport(testInfo) {
+  const width = testInfo.project.use?.viewport?.width || 0;
+  return width > 0 && width <= 768;
+}
+
+/**
+ * Reach Settings through genuine UI controls on either desktop or mobile viewports.
+ * On mobile (<=768px), the rail is hidden and the sidebar must be opened via hamburger.
+ */
+async function openSettings(page, isMobile) {
+  let settingsButton;
+  if (isMobile) {
+    const hamburger = page.locator('#hamburger-btn:visible').first();
+    await expect(hamburger, 'Hamburger button must be visible on mobile').toBeVisible({ timeout: 5000 });
+    await hamburger.click();
+    const sidebar = page.locator('#sidebar:not(.hidden)');
+    await expect(sidebar, 'Sidebar must open upon hamburger tap').toBeVisible({ timeout: 5000 });
+    await page.waitForTimeout(300);
+    settingsButton = sidebar.locator('#user-bar-settings');
+  } else {
+    settingsButton = page.locator('#rail-settings:visible, #user-bar-settings:visible').first();
+  }
+
+  await expect(settingsButton, 'Settings button must be reachable and visible').toBeVisible({ timeout: 5000 });
+  await settingsButton.click();
+  const dialog = page.locator('#settings-modal');
+  await expect(dialog, 'Settings modal must become visible').toBeVisible({ timeout: 5000 });
+  return dialog;
+}
+
+/**
+ * Reach Notes through genuine UI controls on either desktop or mobile viewports.
+ */
+async function openNotes(page, isMobile) {
+  let notesButton;
+  if (isMobile) {
+    const hamburger = page.locator('#hamburger-btn:visible').first();
+    await expect(hamburger, 'Hamburger button must be visible on mobile').toBeVisible({ timeout: 5000 });
+    await hamburger.click();
+    const sidebar = page.locator('#sidebar:not(.hidden)');
+    await expect(sidebar, 'Sidebar must open upon hamburger tap').toBeVisible({ timeout: 5000 });
+    await page.waitForTimeout(300);
+    notesButton = sidebar.locator('#tool-notes-btn');
+  } else {
+    notesButton = page.locator('#rail-notes:visible, #tool-notes-btn:visible').first();
+  }
+
+  await expect(notesButton, 'Notes button must be reachable and visible').toBeVisible({ timeout: 5000 });
+  await notesButton.click();
+  const realNotes = page.locator('.notes-pane').first();
+  await expect(realNotes, 'Notes pane must become visible').toBeVisible({ timeout: 5000 });
+  return realNotes;
+}
+
+/**
+ * Reach Document editor through genuine UI controls on either desktop or mobile viewports.
+ */
+async function openDocument(page, isMobile) {
+  let docButton;
+  if (isMobile) {
+    const hamburger = page.locator('#hamburger-btn:visible').first();
+    await expect(hamburger, 'Hamburger button must be visible on mobile').toBeVisible({ timeout: 5000 });
+    await hamburger.click();
+    const sidebar = page.locator('#sidebar:not(.hidden)');
+    await expect(sidebar, 'Sidebar must open upon hamburger tap').toBeVisible({ timeout: 5000 });
+    await page.waitForTimeout(300);
+    docButton = sidebar.locator('#library-new-doc-btn');
+  } else {
+    docButton = page.locator('#library-new-doc-btn:visible').first();
+    if (!(await docButton.count())) {
+      const hamburger = page.locator('#hamburger-btn:visible').first();
+      if (await hamburger.count()) {
+        await hamburger.click();
+        await page.waitForTimeout(300);
+        docButton = page.locator('#sidebar:not(.hidden) #library-new-doc-btn');
+      }
+    }
+  }
+
+  await expect(docButton, 'Document creation button must be reachable and visible').toBeVisible({ timeout: 5000 });
+  await docButton.click();
+  const docPane = page.locator('.doc-editor-pane').first();
+  await expect(docPane, 'Document editor pane must become visible').toBeVisible({ timeout: 10000 });
+  return docPane;
+}
+
+test('renders actual Odysseus UI and captures review screenshots', async ({ page }, testInfo) => {
+  await openApp(page);
+  const out = testInfo.config.metadata.evidenceDir;
+  fs.mkdirSync(out, { recursive: true });
+  const name = testInfo.project.name;
+  const isMobile = isMobileViewport(testInfo);
+
+  // 1. Chat screenshot (default view)
+  await page.screenshot({ path: path.join(out, `${name}-chat.png`), fullPage: false, animations: 'disabled' });
+
+  // 2. Settings screenshot via actual user-visible navigation
+  await openApp(page);
+  await openSettings(page, isMobile);
+  await page.screenshot({ path: path.join(out, `${name}-settings.png`), fullPage: false, animations: 'disabled' });
+
+  // 3. Notes screenshot via actual user-visible navigation
+  await openApp(page);
+  await openNotes(page, isMobile);
+  await page.screenshot({ path: path.join(out, `${name}-notes.png`), fullPage: false, animations: 'disabled' });
+
+  testInfo.annotations.push({ type: 'evidence', description: out });
+});
+
+test('viewport meta does not restrict user zoom', async ({ page }) => {
+  await openApp(page);
+  const meta = await page.locator('meta[name="viewport"]').getAttribute('content');
+  expect(meta || '').not.toMatch(/maximum-scale\s*=\s*1(?:\b|\.)/i);
+  expect(meta || '').not.toMatch(/user-scalable\s*=\s*no/i);
+});
+
+test('browser-owned window scrolling is not globally reset', async ({ page }) => {
+  await openApp(page);
+  const value = await page.evaluate(async () => {
+    document.documentElement.style.overflowY = 'auto';
+    document.body.style.overflowY = 'auto';
+    document.body.style.height = 'auto';
+    document.body.style.minHeight = '3200px';
+    const spacer = document.createElement('div');
+    spacer.style.cssText = 'position:absolute;top:3100px;width:1px;height:10px;pointer-events:none;';
+    document.body.appendChild(spacer);
+    window.scrollTo(0, 250);
+    await new Promise(resolve => setTimeout(resolve, 150));
+    return window.scrollY;
+  });
+  expect(value).toBeGreaterThan(100);
+});
+
+test('CSS cascade rules for mobile Notes and Documents respect safe areas', async ({ page }, testInfo) => {
+  test.skip(!isMobileViewport(testInfo), 'Only mobile viewports <=768px');
+  await openApp(page);
+  const results = await page.evaluate(() => {
+    document.documentElement.style.setProperty('--safe-top', '44px');
+    document.documentElement.style.setProperty('--sheet-top', '68px');
+    document.documentElement.style.setProperty('--app-height', '844px');
+    const values = {};
+    for (const [key, bodyClass, paneClass] of [
+      ['notes', 'notes-view', 'notes-pane'],
+      ['document', 'doc-view', 'doc-editor-pane'],
+    ]) {
+      const pane = document.createElement('section');
+      pane.className = paneClass;
+      pane.style.opacity = '0'; // CSS cascade regression check
+      document.body.appendChild(pane);
+      document.body.classList.add(bodyClass);
+      const style = window.getComputedStyle(pane);
+      values[key] = {
+        top: style.top,
+        height: style.height,
+        insetTop: pane.getBoundingClientRect().top,
+      };
+      pane.remove();
+      document.body.classList.remove(bodyClass);
+    }
+    return values;
+  });
+  for (const key of ['notes', 'document']) {
+    expect(results[key].top, key).toBe('68px');
+    expect(parseFloat(results[key].height), key).toBeGreaterThanOrEqual(770);
+  }
+});
+
+test('real Notes and Document panes honor safe-area geometry in the live application', async ({ page }, testInfo) => {
+  const isMobile = isMobileViewport(testInfo);
+  test.skip(!isMobile, 'Only mobile viewports <=768px');
+  await openApp(page);
+
+  // Configure simulated safe-area insets (e.g. iPhone notch / Dynamic Island)
+  await page.evaluate(() => {
+    document.documentElement.style.setProperty('--safe-top', '44px');
+    document.documentElement.style.setProperty('--sheet-top', '68px');
+    document.documentElement.style.setProperty('--app-height', '844px');
+  });
+
+  // 1. Verify actual Notes UI
+  const realNotes = await openNotes(page, isMobile);
+  const notesMetrics = await realNotes.evaluate(el => {
+    const style = window.getComputedStyle(el);
+    const rect = el.getBoundingClientRect();
+    return {
+      top: style.top,
+      height: style.height,
+      maxHeight: style.maxHeight,
+      rectTop: rect.top,
+      rectHeight: rect.height,
+    };
+  });
+
+  expect(notesMetrics.top, 'Real Notes pane top must match --sheet-top').toBe('68px');
+  expect(parseFloat(notesMetrics.height), 'Real Notes pane height must respect safe area').toBeGreaterThanOrEqual(770);
+  expect(notesMetrics.rectTop, 'Real Notes pane must not overlap the 44px safe-area top notch').toBeGreaterThanOrEqual(44);
+
+  // 2. Verify actual Document UI
+  await openApp(page);
+  await page.evaluate(() => {
+    document.documentElement.style.setProperty('--safe-top', '44px');
+    document.documentElement.style.setProperty('--sheet-top', '68px');
+    document.documentElement.style.setProperty('--app-height', '844px');
+  });
+
+  const realDoc = await openDocument(page, isMobile);
+  const docMetrics = await realDoc.evaluate(el => {
+    const style = window.getComputedStyle(el);
+    const rect = el.getBoundingClientRect();
+    return {
+      top: style.top,
+      height: style.height,
+      maxHeight: style.maxHeight,
+      rectTop: rect.top,
+      rectHeight: rect.height,
+    };
+  });
+
+  expect(docMetrics.top, 'Real Document pane top must match --sheet-top').toBe('68px');
+  expect(parseFloat(docMetrics.height), 'Real Document pane height must respect safe area').toBeGreaterThanOrEqual(770);
+  expect(docMetrics.rectTop, 'Real Document pane must not overlap the 44px safe-area top notch').toBeGreaterThanOrEqual(44);
+});
+
+test('restored floating windows clamp away from a left safe area', async ({ page }, testInfo) => {
+  test.skip(isMobileViewport(testInfo), 'Geometry test uses desktop resizable windows');
+  await openApp(page);
+  const value = await page.evaluate(async () => {
+    document.documentElement.style.setProperty('--safe-left', '58px');
+    document.body.style.setProperty('padding-left', '58px', 'important');
+    const probe = document.createElement('div');
+    probe.style.cssText = 'position:fixed;left:10px;top:80px;width:420px;height:250px;z-index:999999;';
+    document.body.appendChild(probe);
+    const key = 'odysseus-layout-restore-test';
+    localStorage.setItem(key, JSON.stringify({ w: 480, h: 280 }));
+    try {
+      // Body horizontal padding transitions over 160ms. Wait for the
+      // simulated safe-area change to settle before restoring geometry.
+      await new Promise(resolve => setTimeout(resolve, 300));
+
+      const padAtRestore = parseFloat(
+        getComputedStyle(document.body).paddingLeft
+      );
+
+      if (Math.abs(padAtRestore - 58) > 0.5) {
+        throw new Error(
+          `Safe-area fixture did not settle: ${padAtRestore}px`
+        );
+      }
+
+      // Verify the running server actually serves the patched module.
+      const response = await fetch(
+        '/static/js/windowResize.js',
+        { cache: 'no-store' }
+      );
+      if (!response.ok) {
+        throw new Error('Cannot fetch windowResize.js');
+      }
+      const moduleSource = await response.text();
+      if (!moduleSource.includes('const rightmostLeft = Math.max(padLeft')) {
+        throw new Error('Running server is serving unpatched windowResize.js');
+      }
+
+      const { makeWindowResizable } = await import('/static/js/windowResize.js');
+      makeWindowResizable(probe, { storageKey: key, mobileSkip: 768 });
+      await new Promise(resolve => setTimeout(resolve, 300));
+      return {
+        left: parseFloat(probe.style.left),
+        width: parseFloat(probe.style.width),
+        paddingLeft: parseFloat(getComputedStyle(document.body).paddingLeft),
+        viewportWidth: window.innerWidth,
+      };
+    } finally {
+      localStorage.removeItem(key);
+      probe.remove();
+    }
+  });
+
+  expect(value.paddingLeft).toBeGreaterThanOrEqual(58);
+  expect(value.width).toBe(480);
+  expect(value.left).toBeGreaterThanOrEqual(value.paddingLeft);
+});

tests/e2e/photo-editor/brush-cursor.spec.js

@@ -46,3 +46,480 @@ test('brush cursor matches the rendered brush diameter and replaces the crosshai
   await expect.poll(async () => (await cursor.boundingBox()).width)
     .toBeCloseTo(measurements.expectedWidth * 2, 1);
 });
+
+
+test('brush cursor follows the pointer before, during, and after a stroke and cancels animations on close', async ({ page }) => {
+  await openBlankEditor(page, { width: 400, height: 300 });
+  await selectTool(page, 'brush');
+  const canvas = await page.locator('.ge-main-canvas').boundingBox();
+  const cursor = page.locator('.ge-brush-cursor');
+  const checkCenter = async (x, y) => {
+    await expect.poll(async () => {
+      const box = await cursor.boundingBox();
+      return box ? Math.max(Math.abs(box.x + box.width / 2 - x), Math.abs(box.y + box.height / 2 - y)) : Infinity;
+    }).toBeLessThan(1);
+  };
+  const a = { x: canvas.x + canvas.width * 0.3, y: canvas.y + canvas.height * 0.4 };
+  const b = { x: canvas.x + canvas.width * 0.7, y: canvas.y + canvas.height * 0.6 };
+  await page.mouse.move(a.x, a.y);
+  await checkCenter(a.x, a.y);
+  await page.mouse.down();
+  await page.mouse.move(b.x, b.y, { steps: 8 });
+  await checkCenter(b.x, b.y);
+  await page.mouse.up();
+  await page.mouse.move(a.x, a.y);
+  await checkCenter(a.x, a.y);
+
+  // 1. Ensure at least one animation exists before checking cancellation
+  // 2. Verify the actual cursor-position animation is paused while active
+  const activeAnimation = await page.evaluate(() => {
+    const cursorEl = document.querySelector('.ge-brush-cursor');
+    const anims = cursorEl ? cursorEl.getAnimations() : [];
+    return {
+      count: anims.length,
+      hasPositionAnim: Boolean(cursorEl?._positionAnimation),
+      positionAnimPaused: cursorEl?._positionAnimation?.playState === 'paused',
+      positionAnimCurrentTime: cursorEl?._positionAnimation?.currentTime,
+      allPaused: anims.length > 0 && anims.every(a => a.playState === 'paused'),
+    };
+  });
+  expect(activeAnimation.count).toBeGreaterThanOrEqual(1);
+  expect(activeAnimation.hasPositionAnim).toBe(true);
+  expect(activeAnimation.positionAnimPaused).toBe(true);
+  expect(activeAnimation.positionAnimCurrentTime).toBe(0);
+  expect(activeAnimation.allPaused).toBe(true);
+
+  // 3. Verify it is canceled when the editor closes
+  // 4. Ensure temporary test-only references are cleaned up
+  const closeOutcome = await page.evaluate(async () => {
+    const cursorEl = document.querySelector('.ge-brush-cursor');
+    const anims = cursorEl ? cursorEl.getAnimations() : [];
+    const editor = await window.__photoEditorImport('/static/js/galleryEditor.js');
+    editor.closeEditor({ force: true });
+    return {
+      trackedCount: anims.length,
+      allIdle: anims.length > 0 && anims.every(a => a.playState === 'idle'),
+      states: anims.map(a => a.playState),
+    };
+  });
+  await expect(cursor).toHaveCount(0);
+  expect(closeOutcome.trackedCount).toBeGreaterThanOrEqual(1);
+  expect(closeOutcome.allIdle).toBe(true);
+  expect(closeOutcome.states.every(s => s === 'idle')).toBe(true);
+
+  // Verify no temporary references linger on window
+  const windowClean = await page.evaluate(() => {
+    return !('__cursorAnimationsForTest' in window) && !('__styleTransformMutations' in window);
+  });
+  expect(windowClean).toBe(true);
+});
+
+
+test('steady pointer movement does not perform repeated inline style-transform writes', async ({ page }) => {
+  await openBlankEditor(page, { width: 400, height: 300 });
+  await selectTool(page, 'brush');
+
+  const canvas = await page.locator('.ge-main-canvas').boundingBox();
+  const cursor = page.locator('.ge-brush-cursor');
+
+  // Warm up the cursor first: move pointer onto canvas
+  const startX = canvas.x + canvas.width * 0.25;
+  const startY = canvas.y + canvas.height * 0.25;
+  await page.mouse.move(startX, startY);
+  await expect(cursor).toBeVisible();
+
+  // Observe style attribute mutations during steady pointer movement.
+  // Brush size, color, and mode remain unchanged throughout.
+  await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    window.__styleMutationRecords = [];
+    const observer = new MutationObserver((mutations) => {
+      for (const m of mutations) {
+        if (m.type === 'attributes' && m.attributeName === 'style') {
+          const oldStyle = m.oldValue || '';
+          const currentStyle = el.getAttribute('style') || '';
+          // Track mutations where style-transform was written inline
+          if (el.style.transform || currentStyle.includes('transform:') || oldStyle.includes('transform:')) {
+            window.__styleMutationRecords.push({
+              oldStyle,
+              currentStyle,
+              inlineTransform: el.style.transform,
+            });
+          }
+        }
+      }
+    });
+    observer.observe(el, { attributes: true, attributeFilter: ['style'], attributeOldValue: true });
+    window.__styleMutationObserver = observer;
+  });
+
+  // Preserve actual mouse interactions: steady movement across 10 steps
+  const endX = canvas.x + canvas.width * 0.75;
+  const endY = canvas.y + canvas.height * 0.75;
+  await page.mouse.move(endX, endY, { steps: 10 });
+
+  // Confirm the cursor accurately tracked the pointer
+  await expect.poll(async () => {
+    const box = await cursor.boundingBox();
+    return box ? Math.max(Math.abs(box.x + box.width / 2 - endX), Math.abs(box.y + box.height / 2 - endY)) : Infinity;
+  }).toBeLessThan(1);
+
+  // Collect observation results and clean up observer and globals
+  const { mutationCount, inlineTransform, animCount } = await page.evaluate(() => {
+    window.__styleMutationObserver?.disconnect();
+    delete window.__styleMutationObserver;
+    const records = window.__styleMutationRecords || [];
+    delete window.__styleMutationRecords;
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      mutationCount: records.length,
+      inlineTransform: el?.style.transform || '',
+      animCount: el?.getAnimations?.().length || 0,
+    };
+  });
+
+  // The optimized implementation uses paused Web Animations keyframes, avoiding repeated
+  // inline style-transform writes that trigger expensive :has() document invalidations.
+  expect(mutationCount).toBe(0);
+  expect(inlineTransform).toBe('');
+  expect(animCount).toBeGreaterThanOrEqual(1);
+});
+
+
+test('cross-tool cursor positioning, sizing, and styling across brush, eraser, pixel, lasso, inpaint, liquify, and clone', async ({ page }) => {
+  await openBlankEditor(page, { width: 400, height: 300 });
+  const canvas = await page.locator('.ge-main-canvas').boundingBox();
+  const cursor = page.locator('.ge-brush-cursor');
+
+  const checkCenter = async (x, y) => {
+    await expect.poll(async () => {
+      const box = await cursor.boundingBox();
+      return box ? Math.max(Math.abs(box.x + box.width / 2 - x), Math.abs(box.y + box.height / 2 - y)) : Infinity;
+    }).toBeLessThan(1);
+  };
+
+  const midX = canvas.x + canvas.width / 2;
+  const midY = canvas.y + canvas.height / 2;
+
+  // 1. Brush tool
+  await selectTool(page, 'brush');
+  await page.mouse.move(midX, midY);
+  await expect(cursor).toBeVisible();
+  await checkCenter(midX, midY);
+  let props = await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      isPixel: el.classList.contains('ge-brush-cursor--pixel'),
+      display: el.style.display,
+      borderStyle: el.style.borderStyle,
+    };
+  });
+  expect(props.isPixel).toBe(false);
+  expect(props.display).toBe('block');
+
+  // 2. Eraser tool
+  await selectTool(page, 'eraser');
+  await page.mouse.move(midX + 10, midY + 10);
+  await checkCenter(midX + 10, midY + 10);
+  props = await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      borderColor: el.style.borderColor,
+      isPixel: el.classList.contains('ge-brush-cursor--pixel'),
+    };
+  });
+  expect(props.borderColor).toBe('rgba(255, 255, 255, 0.6)');
+  expect(props.isPixel).toBe(false);
+
+  // 3. Pixel brush (via tool / preset)
+  await page.evaluate(async () => {
+    const { state } = await import('/static/js/editor/state.js');
+    state.tool = 'pixel';
+    state.brushPixel = true;
+    state.brushSize = 16;
+  });
+  await page.mouse.move(midX - 10, midY - 10);
+  await checkCenter(midX - 10, midY - 10);
+  props = await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      isPixel: el.classList.contains('ge-brush-cursor--pixel'),
+      width: parseFloat(el.style.width),
+    };
+  });
+  expect(props.isPixel).toBe(true);
+
+  // 4. Lasso tool
+  await page.evaluate(async () => {
+    const { state } = await import('/static/js/editor/state.js');
+    state.tool = 'lasso';
+    state.brushPixel = false;
+  });
+  await page.mouse.move(midX + 20, midY - 10);
+  await checkCenter(midX + 20, midY - 10);
+  props = await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      borderColor: el.style.borderColor,
+      background: el.style.background,
+    };
+  });
+  expect(props.borderColor).toBe('rgba(255, 255, 255, 0.85)');
+  expect(props.background).toBe('rgba(0, 0, 0, 0.15)');
+
+  // 5. Inpaint tool (paint mode and erase mode)
+  await page.evaluate(async () => {
+    const { state } = await import('/static/js/editor/state.js');
+    state.tool = 'inpaint';
+    state.inpaintEraseMode = false;
+  });
+  await page.mouse.move(midX, midY);
+  await checkCenter(midX, midY);
+  props = await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      borderColor: el.style.borderColor,
+      background: el.style.background,
+      borderStyle: el.style.borderStyle,
+    };
+  });
+  expect(props.borderColor).toBe('rgba(255, 80, 80, 0.8)');
+  expect(props.borderStyle).toBe('solid');
+
+  // Inpaint erase mode
+  await page.evaluate(async () => {
+    const { state } = await import('/static/js/editor/state.js');
+    state.inpaintEraseMode = true;
+  });
+  await page.mouse.move(midX + 5, midY + 5);
+  props = await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      borderColor: el.style.borderColor,
+      borderStyle: el.style.borderStyle,
+    };
+  });
+  expect(props.borderColor).toBe('rgba(255, 255, 255, 0.9)');
+  expect(props.borderStyle).toBe('dashed');
+
+  // 6. Liquify tool
+  await page.evaluate(async () => {
+    const { state } = await import('/static/js/editor/state.js');
+    state.tool = 'liquify';
+    state.liquifySize = 100;
+  });
+  await page.mouse.move(midX - 15, midY + 15);
+  await checkCenter(midX - 15, midY + 15);
+  props = await page.evaluate(() => {
+    const el = document.querySelector('.ge-brush-cursor');
+    return {
+      width: parseFloat(el.style.width),
+      height: parseFloat(el.style.height),
+      display: el.style.display,
+    };
+  });
+  expect(props.display).toBe('block');
+  expect(props.width).toBeGreaterThan(0);
+
+  // 7. Clone tool
+  await page.evaluate(async () => {
+    const { state } = await import('/static/js/editor/state.js');
+    state.tool = 'clone';
+    state.brushSize = 50;
+  });
+  await page.mouse.move(midX, midY);
+  await checkCenter(midX, midY);
+  await expect(cursor).toBeVisible();
+
+  // Clean close
+  await page.evaluate(async () => {
+    const editor = await window.__photoEditorImport('/static/js/galleryEditor.js');
+    editor.closeEditor({ force: true });
+  });
+  await expect(cursor).toHaveCount(0);
+});
+
+
+test('canvas boundaries, pointer exit and re-entry, and viewport resize', async ({ page }) => {
+  await openBlankEditor(page, { width: 400, height: 300 });
+  await selectTool(page, 'brush');
+
+  const canvas = await page.locator('.ge-main-canvas').boundingBox();
+  const cursor = page.locator('.ge-brush-cursor');
+
+  const checkCenter = async (x, y) => {
+    await expect.poll(async () => {
+      const box = await cursor.boundingBox();
+      return box ? Math.max(Math.abs(box.x + box.width / 2 - x), Math.abs(box.y + box.height / 2 - y)) : Infinity;
+    }).toBeLessThan(1);
+  };
+
+  // 1. Enter canvas
+  const insideX = canvas.x + canvas.width * 0.5;
+  const insideY = canvas.y + canvas.height * 0.5;
+  await page.mouse.move(insideX, insideY);
+  await expect(cursor).toBeVisible();
+  await checkCenter(insideX, insideY);
+
+  // 2. Exit canvas boundary: cursor should hide
+  const outsideX = canvas.x - 50;
+  const outsideY = canvas.y - 50;
+  await page.mouse.move(outsideX, outsideY);
+  await expect(cursor).toBeHidden();
+
+  // 3. Re-enter canvas: cursor should restore visibility and position
+  await page.mouse.move(insideX, insideY);
+  await expect(cursor).toBeVisible();
+  await checkCenter(insideX, insideY);
+
+  // 4. Viewport resize
+  await page.setViewportSize({ width: 1280, height: 800 });
+  await page.waitForTimeout(100);
+  const newCanvas = await page.locator('.ge-main-canvas').boundingBox();
+  const newInsideX = newCanvas.x + newCanvas.width * 0.5;
+  const newInsideY = newCanvas.y + newCanvas.height * 0.5;
+  await page.mouse.move(newInsideX, newInsideY);
+  await expect(cursor).toBeVisible();
+  await checkCenter(newInsideX, newInsideY);
+
+  // Clean close
+  await page.evaluate(async () => {
+    const editor = await window.__photoEditorImport('/static/js/galleryEditor.js');
+    editor.closeEditor({ force: true });
+  });
+  await expect(cursor).toHaveCount(0);
+});
+
+
+for (const view of ['2x zoom', 'fit-to-view']) {
+  test(`zoom and fit-to-view cursor scaling and positioning: ${view}`, async ({ page }) => {
+    const size = view === '2x zoom' ? { width: 400, height: 300 } : { width: 1600, height: 1200 };
+    await openBlankEditor(page, size);
+    await selectTool(page, 'brush');
+
+    await page.evaluate(async () => {
+      const { state } = await import('/static/js/editor/state.js');
+      state.brushSize = 50;
+      // The zoom-in control applies 1.25x through the real zoom handler.
+      state.zoom = 1.6;
+      state.zoomViewMode = 'scale';
+    });
+    await page.locator('#ge-zoom-in').click();
+    await expect.poll(() => page.evaluate(async () => {
+      const { state } = await import('/static/js/editor/state.js');
+      return { zoom: state.zoom, mode: state.zoomViewMode };
+    })).toEqual({ zoom: 2, mode: 'scale' });
+
+    if (view === 'fit-to-view') {
+      await page.locator('#ge-zoom-100').click();
+      const fitted = await page.evaluate(async () => {
+        const { state } = await import('/static/js/editor/state.js');
+        const area = document.querySelector('.ge-canvas-area');
+        const pad = state._footerDocked ? 0 : 20;
+        const rulerPad = state.rulersVisible ? 18 : 0;
+        return {
+          zoom: state.zoom,
+          mode: state.zoomViewMode,
+          expectedZoom: Math.min(1,
+            (area.clientWidth - 2 * pad - rulerPad) / state.imgWidth,
+            (area.clientHeight - 2 * pad - rulerPad) / state.imgHeight),
+        };
+      });
+      expect(fitted.mode).toBe('fit');
+      expect(fitted.zoom).toBeGreaterThan(0);
+      expect(fitted.zoom).toBeLessThan(1);
+      expect(fitted.zoom).toBeCloseTo(fitted.expectedZoom, 6);
+    }
+
+    const canvas = page.locator('.ge-main-canvas');
+    const canvasBox = await canvas.boundingBox();
+    const cursor = page.locator('.ge-brush-cursor');
+    if (view === '2x zoom') {
+      expect(canvasBox.width).toBeCloseTo(size.width * 2, 1);
+      expect(canvasBox.height).toBeCloseTo(size.height * 2, 1);
+    }
+    const targetX = canvasBox.x + canvasBox.width * 0.4;
+    const targetY = canvasBox.y + canvasBox.height * 0.4;
+    await page.mouse.move(targetX, targetY);
+    await expect(cursor).toBeVisible();
+
+    // Pointer coordinates can be rounded to device pixels; keep the existing
+    // subpixel alignment tolerance separate from the stricter diameter check.
+    await expect.poll(async () => {
+      const box = await cursor.boundingBox();
+      return box ? Math.max(
+        Math.abs(box.x + box.width / 2 - targetX),
+        Math.abs(box.y + box.height / 2 - targetY),
+      ) : Infinity;
+    }).toBeLessThan(1);
+    await expect.poll(() => page.evaluate(() => {
+      const canvasEl = document.querySelector('.ge-main-canvas');
+      const canvasRect = canvasEl.getBoundingClientRect();
+      const cursorRect = document.querySelector('.ge-brush-cursor').getBoundingClientRect();
+      return Math.max(
+        Math.abs(cursorRect.width - 50 * canvasRect.width / canvasEl.width),
+        Math.abs(cursorRect.height - 50 * canvasRect.height / canvasEl.height),
+      );
+    })).toBeLessThan(0.1);
+
+    await page.evaluate(async () => {
+      const editor = await window.__photoEditorImport('/static/js/galleryEditor.js');
+      editor.closeEditor({ force: true });
+    });
+    await expect(cursor).toHaveCount(0);
+  });
+}
+
+
+test('memory and lifecycle: repeated editor open/close cycles retain no detached cursors or active animations', async ({ page }) => {
+  for (let cycle = 0; cycle < 5; cycle++) {
+    await openBlankEditor(page, { width: 400, height: 300 });
+    await selectTool(page, 'brush');
+
+    const canvas = await page.locator('.ge-main-canvas').boundingBox();
+    await page.mouse.move(canvas.x + canvas.width * 0.5, canvas.y + canvas.height * 0.5);
+    const cursor = page.locator('.ge-brush-cursor');
+    await expect(cursor).toBeVisible();
+
+    // Verify cursor layer and animation are active
+    const cycleState = await page.evaluate(() => {
+      const el = document.querySelector('.ge-brush-cursor');
+      window.__lastCycleAnim = el?._positionAnimation;
+      return {
+        cursorCount: document.querySelectorAll('.ge-brush-cursor').length,
+        layerCount: document.querySelectorAll('#ge-cursor-layer').length,
+        hasAnim: Boolean(el?._positionAnimation),
+        animPaused: el?._positionAnimation?.playState === 'paused',
+      };
+    });
+    expect(cycleState.cursorCount).toBe(1);
+    expect(cycleState.layerCount).toBe(1);
+    expect(cycleState.animPaused).toBe(true);
+
+    // Close editor
+    await page.evaluate(async () => {
+      const editor = await window.__photoEditorImport('/static/js/galleryEditor.js');
+      editor.closeEditor({ force: true });
+    });
+
+    // Verify complete cleanup after close
+    await expect(cursor).toHaveCount(0);
+    const postClose = await page.evaluate(() => {
+      const anim = window.__lastCycleAnim;
+      delete window.__lastCycleAnim;
+      return {
+        cursorCount: document.querySelectorAll('.ge-brush-cursor').length,
+        layerCount: document.querySelectorAll('#ge-cursor-layer').length,
+        animIdle: anim ? anim.playState === 'idle' : true,
+        activeAnims: document.getAnimations().filter(a => {
+          const target = a.effect?.target;
+          return target && (target.classList?.contains('ge-brush-cursor') || target.id === 'ge-cursor-layer');
+        }).length,
+      };
+    });
+    expect(postClose.cursorCount).toBe(0);
+    expect(postClose.layerCount).toBe(0);
+    expect(postClose.animIdle).toBe(true);
+    expect(postClose.activeAnims).toBe(0);
+  }
+});

tests/e2e/photo-editor/brush-engine.spec.js

@@ -35,7 +35,7 @@ test('brush presets and shared stroke controls remain reusable', async ({ page }
   await expect(page.locator('#ge-brush-preset option', { hasText: 'My Detail Brush' })).toHaveCount(0);
 });
 
-test('long sampled strokes stay responsive and undo atomically', async ({ page }) => {
+test('long sampled strokes stay responsive and undo atomically', async ({ page, browserName }, testInfo) => {
   await openBlankEditor(page);
   await selectTool(page, 'brush');
   await page.locator('.ge-size-slider').fill('700');
@@ -49,13 +49,40 @@ test('long sampled strokes stay responsive and undo atomically', async ({ page }
   const before = await flattenedPixelDigest(page);
   const start = await canvasPoint(page, 0.08, 0.5);
   const end = await canvasPoint(page, 0.92, 0.5);
+  // Playwright delivers stepped moves at frame cadence: 240 events already
+  // cost ~4s on an inert page. Measure that transport cost with the same input,
+  // and separately bound real Chromium main-thread work at the original5s.
+  const control = await page.context().newPage();
+  let controlDuration;
+  try {
+    await control.setContent('<canvas style="position:fixed;inset:0;width:100vw;height:100vh"></canvas>');
+    const started = Date.now();
+    await control.mouse.move(start.x, start.y);
+    await control.mouse.down();
+    await control.mouse.move(end.x, end.y, { steps: 240 });
+    await control.mouse.up();
+    controlDuration = Date.now() - started;
+  } finally {
+    await control.close();
+    await page.bringToFront();
+  }
+  const metrics = browserName === 'chromium' ? await page.context().newCDPSession(page) : null;
+  if (metrics) await metrics.send('Performance.enable');
+  const taskTime = async () => (await metrics.send('Performance.getMetrics')).metrics.find(item => item.name === 'TaskDuration').value *1000;
+  const beforeTask = metrics ? await taskTime() : null;
   const startedAt = Date.now();
   await page.mouse.move(start.x, start.y);
   await page.mouse.down();
   await page.mouse.move(end.x, end.y, { steps: 240 });
   await page.mouse.up();
   const strokeDuration = Date.now() - startedAt;
-  expect(strokeDuration).toBeLessThan(5000);
+  const mainThreadDuration = metrics ? (await taskTime()) - beforeTask : null;
+  if (metrics) await metrics.detach();
+  await testInfo.attach('stroke-timing', { body: Buffer.from(JSON.stringify({
+    samples: 240, strokeDuration, controlDuration, mainThreadDuration,
+  })), contentType: 'application/json' });
+  expect(strokeDuration - controlDuration).toBeLessThan(5000);
+  if (mainThreadDuration !== null) expect(mainThreadDuration).toBeLessThan(5000);
   const after = await flattenedPixelDigest(page);
   expect(after).not.toEqual(before);
   await expect(thumb).toHaveAttribute('width', '68');

tests/e2e/photo-editor/email-reply-native.spec.js

@@ -0,0 +1,21 @@
+const { test, expect } = require('@playwright/test');
+const { spawnSync } = require('node:child_process');
+const path = require('node:path');
+
+// The previously opt-in node:test now uses only this config's owned app/account.
+// Its native JUnit report retains its authored identity; this case is a driver.
+test('native email reply regression on the disposable authenticated app', async ({ context, baseURL }, testInfo) => {
+  const origin = new URL(baseURL);
+  expect(origin.hostname).toBe('127.0.0.1');
+  const cookie = (await context.cookies(baseURL)).find(item => item.name === 'odysseus_session');
+  expect(cookie?.value).toBeTruthy();
+  const report = testInfo.outputPath('native-email-junit.xml');
+  const result = spawnSync(process.execPath, ['--test', '--test-reporter=spec', '--test-reporter-destination=stdout',
+    '--test-reporter=junit', `--test-reporter-destination=${report}`, 'tests/emailReplyBrowser.test.mjs'], {
+    cwd: path.resolve(__dirname, '../../..'), encoding: 'utf8', timeout: 80_000,
+    env: { ...process.env, ODYSSEUS_UI_TEST_URL: baseURL, ODYSSEUS_UI_TEST_COOKIE: cookie.value },
+  });
+  console.log(result.stdout);
+  expect(result.error, String(result.error)).toBeUndefined();
+  expect(result.status, result.stdout + result.stderr).toBe(0);
+});

tests/e2e/playwright.config.js

@@ -2,6 +2,12 @@ const fs = require('node:fs');
 const os = require('node:os');
 const path = require('node:path');
 const { defineConfig } = require('@playwright/test');
+const { disposableCIArtifacts, marker } = require('./artifact-policy');
+const isCI = Boolean(process.env.CI) || Boolean(process.env.GITHUB_ACTIONS);
+const disposableOptIn = isCI && process.env.ODYSSEUS_E2E_DISPOSABLE_ARTIFACTS === '1';
+if (disposableOptIn && (process.env.PHOTO_EDITOR_E2E_DATA_DIR || process.env.PHOTO_EDITOR_E2E_DB_PATH || process.env.PHOTO_EDITOR_E2E_AUTH === '0') && !disposableCIArtifacts()) {
+  throw new Error('CI artifact opt-in requires fresh authenticated fixture paths');
+}
 
 const port = Number(process.env.PHOTO_EDITOR_E2E_PORT || 7013);
 const baseURL = `http://127.0.0.1:${port}`;
@@ -24,7 +30,25 @@ if (!['chromium', 'firefox', 'webkit'].includes(browserName)) {
   throw new Error(`Unsupported PHOTO_EDITOR_E2E_BROWSER: ${browserName}`);
 }
 
-const isCI = Boolean(process.env.CI) || Boolean(process.env.GITHUB_ACTIONS);
+if (disposableOptIn) {
+  const fixturePayload = JSON.stringify({ synthetic: true, authenticated: true, database: databasePath });
+  fs.writeFileSync(path.join(dataDirectory, marker), fixturePayload);
+  const resultsDir = path.join(repositoryRoot, 'test-results');
+  if (!fs.existsSync(resultsDir)) {
+    try { fs.mkdirSync(resultsDir, { recursive: true }); } catch (_) {}
+  }
+  try {
+    fs.writeFileSync(path.join(resultsDir, marker), fixturePayload);
+  } catch (_) {}
+  if (process.env.GITHUB_ENV) {
+    try {
+      fs.appendFileSync(
+        process.env.GITHUB_ENV,
+        `PHOTO_EDITOR_E2E_DATA_DIR=${dataDirectory}\nPHOTO_EDITOR_E2E_DB_PATH=${databasePath}\n`
+      );
+    } catch (_) {}
+  }
+}
 const traceOptIn = !isCI && ['1', 'true'].includes(String(process.env.PLAYWRIGHT_ENABLE_TRACE || '').toLowerCase());
 
 module.exports = defineConfig({
@@ -32,6 +56,8 @@ module.exports = defineConfig({
   outputDir: path.join(__dirname, '..', '..', 'test-results', 'photo-editor'),
   timeout: 90_000,
   expect: { timeout: 10_000 },
+  forbidOnly: isCI,
+  retries: 0,
   fullyParallel: false,
   workers: 1,
   reporter: isCI ? 'line' : 'line',
@@ -49,9 +75,9 @@ module.exports = defineConfig({
     serviceWorkers: 'block',
     ...(authenticated ? { storageState: path.join(dataDirectory, 'browser-auth.json') } : {}),
     screenshot: 'only-on-failure',
-    // Unconditional CI trace suppression: Playwright traces are NEVER recorded in CI.
+    // Default CI suppression remains; only the fresh synthetic fixture opts in.
     // Local developer tracing requires explicit PLAYWRIGHT_ENABLE_TRACE=1.
-    trace: traceOptIn ? 'retain-on-failure' : 'off',
+    trace: (traceOptIn || disposableCIArtifacts()) ? 'retain-on-failure' : 'off',
   },
   webServer: {
     command: `${JSON.stringify(python)} -m uvicorn app:app --host 127.0.0.1 --port ${port}`,
@@ -65,6 +91,10 @@ module.exports = defineConfig({
       DATABASE_URL: `sqlite:///${databasePath}`,
       ODYSSEUS_DATA_DIR: dataDirectory,
       ODYSSEUS_STARTUP_WARMUPS: '0',
+      ODYSSEUS_TOOL_INDEX_PREWARM: '0',
+      ODYSSEUS_BROWSER_MCP_REQUIRE_CACHE: '1',
+      npm_config_offline: 'true',
+      npm_config_cache: path.join(dataDirectory, 'npm-cache'),
       RESPONSE_COMPRESSION_ENABLED: 'false',
     },
   },

tests/e2e/teardown.js

@@ -2,6 +2,7 @@ const fs = require('node:fs');
 const os = require('node:os');
 const path = require('node:path');
 const { execFileSync } = require('node:child_process');
+const { disposableCIArtifacts, marker } = require('./artifact-policy');
 
 const repoRoot = path.resolve(__dirname, '..', '..');
 const defaultResultsDir = path.join(repoRoot, 'test-results');
@@ -29,6 +30,9 @@ module.exports = async (options = {}) => {
   const isPlaywrightConfig = options && (options.configFile || options.rootDir);
   const testOpts = (!isPlaywrightConfig && typeof options === 'object') ? options : {};
 
+  // Validate fixture ownership before removing its temporary directory.
+  const disposableCI = disposableCIArtifacts();
+
   // Confine cleanup of PHOTO_EDITOR_E2E_DATA_DIR:
   // Strictly require it to be within an OS temporary directory and match the expected test prefix
   const dataDirectory = process.env.PHOTO_EDITOR_E2E_DATA_DIR;
@@ -100,7 +104,7 @@ module.exports = async (options = {}) => {
     const isCI = Boolean(process.env.CI) || Boolean(process.env.GITHUB_ACTIONS);
     const traceOptIn = !isCI && ['1', 'true'].includes(String(process.env.PLAYWRIGHT_ENABLE_TRACE || '').toLowerCase());
 
-    if (isCI || !traceOptIn) {
+    if ((isCI && !disposableCI) || (!traceOptIn && !disposableCI)) {
       // In CI or default local runs, purge all sensitive trace archives and artifacts
       purgeSensitiveArtifacts(resultsDir);
     } else {
@@ -124,7 +128,17 @@ module.exports = async (options = {}) => {
 
       if (!sanitizerSucceeded) {
         // Fail closed: purge all remaining zip or tmp archives
-        purgeZipAndTmp(resultsDir);
+        if (disposableCI) purgeSensitiveArtifacts(resultsDir);
+        else purgeZipAndTmp(resultsDir);
+      } else if (disposableCI) {
+        // Re-affirm fixture proof in resultsDir so independent verifier can establish ownership
+        try {
+          fs.writeFileSync(path.join(resultsDir, marker), JSON.stringify({
+            synthetic: true,
+            authenticated: true,
+            database: process.env.PHOTO_EDITOR_E2E_DB_PATH || path.join(dataDirectory || '', 'app.db'),
+          }));
+        } catch (_) {}
       }
     }
   }

tests/native-ci-manifest.json

@@ -0,0 +1,71 @@
+{
+  "native-ci": [
+    "tests/backgroundToolJobs.test.mjs",
+    "tests/chatEditorProgress.test.mjs",
+    "tests/chatImageDeletion.test.mjs",
+    "tests/chatPhotoEditorIntegrity.test.mjs",
+    "tests/chatProcessingHandoff.test.mjs",
+    "tests/copyChatHeader.test.mjs",
+    "tests/emailReplyStream.test.mjs",
+    "tests/galleryModelPickerSync.test.mjs",
+    "tests/generatedImageResult.test.mjs",
+    "tests/researchMobileButtons.test.mjs",
+    "tests/schemaThinkingProbe.test.mjs",
+    "tests/sidebarNewChat.test.mjs",
+    "tests/skillsApproval.test.mjs",
+    "tests/streamingRendererRaf.test.mjs",
+    "tests/toolFollowupOracle.test.mjs",
+    "tests/tool_followup_oracle.test.mjs"
+  ],
+  "browser-ci": [
+    "tests/emailReplyBrowser.test.mjs"
+  ],
+  "python-ci": {
+    "tests/historyResumeRendering.test.mjs": [
+      "tests/test_history_resume_rendering_js.py",
+      "tests/test_live_fallback_round_attribution.py"
+    ],
+    "tests/live_thinking_scheduler.test.mjs": [
+      "tests/test_live_thinking_scheduler_js.py"
+    ],
+    "tests/noteTestOracle.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/notesApplication.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/notesCharacterization.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/notesDraftAutosave.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/notesInfrastructure.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/notesRecurrence.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/notesReminders.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/notesWorkflows.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/selectedNotesContext.test.mjs": [
+      "tests/test_notes_reference_contracts_js.py"
+    ],
+    "tests/streaming/invariant.test.mjs": [
+      "tests/test_streaming_segmenter_js.py"
+    ],
+    "tests/streaming/segmenter.test.mjs": [
+      "tests/test_streaming_segmenter_js.py"
+    ],
+    "tests/turnRendering.test.mjs": [
+      "tests/test_turn_rendering_js.py"
+    ]
+  },
+  "architecture-ci": [
+    "tests/frontendArchitecture.test.mjs"
+  ]
+}

tests/test_admin_listing_authorization.py

@@ -14,12 +14,22 @@ from routes.auth_routes import SESSION_COOKIE
 from tests.helpers.database import disposable_database
 
 
-@pytest.fixture
+@pytest.fixture(scope="module")
-def authenticated_app(tmp_path, monkeypatch):
+def auth_configuration(tmp_path_factory):
-    monkeypatch.setenv("AUTH_ENABLED", "true")
+    """Hash real fixture passwords once; only immutable bytes are reused."""
-    auth = AuthManager(str(tmp_path / "auth.json"))
+    path = tmp_path_factory.mktemp("admin-auth-seed") / "auth.json"
+    auth = AuthManager(str(path))
     for user, admin in (("member", False), ("admin", True)):
         assert auth.create_user(user, "disposable-test-password", is_admin=admin)
+    return path.read_bytes()
+
+
+@pytest.fixture
+def authenticated_app(tmp_path, monkeypatch, auth_configuration):
+    monkeypatch.setenv("AUTH_ENABLED", "true")
+    path = tmp_path / "auth.json"
+    path.write_bytes(auth_configuration)
+    auth = AuthManager(str(path))
     app = FastAPI()
     app.state.auth_manager = auth
 
@@ -133,3 +143,17 @@ def test_diagnostics_require_real_admin_before_sensitive_work(
         for operation in (stats, rag.get_stats, research.call_research_service,
                           extract_id, transcript, collect, read_log):
             operation.assert_not_called()
+
+
+@pytest.mark.parametrize("instance", [0, 1])
+def test_authenticated_fixture_has_private_users_sessions_and_file(authenticated_app, instance):
+    app, client_for = authenticated_app
+    auth = app.state.auth_manager
+    assert set(auth.users) == {"member", "admin"}
+    assert not auth._sessions
+    with client_for("member") as client:
+        assert auth.get_username_for_token(client.cookies.get(SESSION_COOKIE)) == "member"
+    assert auth.create_user("fixture-local", "disposable-test-password")
+    # Both parameter cases mutate their own file. Any shared manager, mutable
+    # users/sessions or written-back seed makes the next case fail, in any order.
+    assert "fixture-local" in auth.users

tests/test_agent_editor_routes.py

@@ -1929,23 +1929,60 @@ def test_place_layer_reports_false_on_queued_response(monkeypatch):
     assert placed is False
 
 
-def test_bridge_await_cancellation_handled_cleanly():
+def test_bridge_await_cancellation_handled_cleanly(monkeypatch):
     import asyncio
+    from contextlib import suppress
     from httpx import ASGITransport, AsyncClient
     app = _bridge_app()
 
     async def scenario():
+        entered_wait = asyncio.Event()
+        real_wait_for = asyncio.wait_for
+        waits = []
+
+        async def observe_wait(awaitable, timeout):
+            # Observe the real wait; do not replace its result or cancellation.
+            waits.append(awaitable)
+            entered_wait.set()
+            return await real_wait_for(awaitable, timeout)
+
+        monkeypatch.setattr(eaar.asyncio, "wait_for", observe_wait)
         async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as c:
             cmd_resp = await c.post("/api/editor/bridge/command", json={"session": "cancel-sess", "op": "document.info", "wait": False})
+            assert cmd_resp.status_code == 200
+            assert cmd_resp.json()["completed"] is False
             cid = cmd_resp.json()["queued"][0]
 
             await_task = asyncio.create_task(c.post("/api/editor/bridge/await", json={"id": cid, "timeout": 30}))
-            await asyncio.sleep(0.05)
-            await_task.cancel()
             try:
-                await await_task
+                await real_wait_for(entered_wait.wait(), 5)
-            except asyncio.CancelledError:
+                assert len(waits) == 1
-                pass
+                assert not waits[0].done()
+                assert not await_task.done(), "await returned without a queued result"
+                await_task.cancel()
+                response = await real_wait_for(await_task, 5)
+                assert response.status_code == 200
+                assert response.json() == {
+                    "ok": False, "error": "cancelled", "status": "cancelled",
+                    "completed": False, "id": cid,
+                }
+            finally:
+                if not await_task.done():
+                    await_task.cancel()
+                    with suppress(asyncio.CancelledError):
+                        await await_task
+
+            # Cancelling the waiter must not cancel the queued command itself.
+            payload = {"width": 137, "height": 251}
+            delivered = await c.post("/api/editor/bridge/result", json={"id": cid, "ok": True, "result": payload})
+            assert delivered.status_code == 200
+            completed = await c.post("/api/editor/bridge/await", json={"id": cid, "timeout": 1})
+            assert completed.status_code == 200
+            assert completed.json()["status"] == "completed"
+            assert completed.json()["completed"] is True
+            assert completed.json()["ok"] is True
+            assert completed.json()["result"] == payload
+            assert len(waits) == 2
 
     asyncio.run(scenario())
 

tests/test_auth_config_lock_concurrency.py

@@ -206,35 +206,50 @@ class TestDiskConsistency:
     @pytest.mark.slow
     def test_file_always_valid_json_during_concurrent_ops(self, tmp_path):
         mgr = _fresh_auth_manager(tmp_path)
-        mgr.create_user("admin", "adminpw", is_admin=True)
+        assert mgr.create_user("admin", "adminpw", is_admin=True)
 
         stop_event = threading.Event()
+        reader_ready = threading.Event()
         corruption_found = []
+        observations = {"reads": 0, "with_writes": 0}
 
         def reader():
             while not stop_event.is_set():
                 try:
                     with open(mgr.auth_path, "r") as f:
                         content = f.read()
-                    json.loads(content)
+                    data = json.loads(content)
+                    observations["reads"] += 1
+                    if any(name.startswith("stress") for name in data["users"]):
+                        observations["with_writes"] += 1
+                    reader_ready.set()
                 except json.JSONDecodeError as e:
                     corruption_found.append(str(e))
                     break
-                except FileNotFoundError:
-                    pass
                 time.sleep(0.001)
 
         def writer():
+            assert reader_ready.wait(5), "reader never observed the initial file"
+            results = []
             for i in range(50):
-                mgr.create_user(f"stress{i}", f"pw{i}")
+                results.append(mgr.create_user(f"stress{i}", f"pw{i}"))
+            return results
 
-        reader_thread = threading.Thread(target=reader)
+        # Futures propagate reader/writer exceptions to the test thread.
-        writer_thread = threading.Thread(target=writer)
+        with ThreadPoolExecutor(max_workers=2) as pool:
-
+            reading = pool.submit(reader)
-        reader_thread.start()
+            writing = pool.submit(writer)
-        writer_thread.start()
+            try:
-        writer_thread.join()
+                results = writing.result(timeout=120)
-        stop_event.set()
+            finally:
-        reader_thread.join()
+                stop_event.set()
+            reading.result(timeout=5)
 
         assert not corruption_found, f"Corrupt JSON detected: {corruption_found[0]}"
+        assert len(results) == 50 and all(results), "writer operations did not succeed"
+        assert observations["reads"] > 0
+        assert observations["with_writes"] > 0, "reader saw no persisted writer activity"
+        expected = {"admin"} | {f"stress{i}" for i in range(50)}
+        with open(mgr.auth_path) as persisted:
+            assert set(json.load(persisted)["users"]) == expected
+        assert set(_fresh_auth_manager(tmp_path).users) == expected

tests/test_caldav_redirect_hardening.py

@@ -13,6 +13,7 @@ import socketserver
 import threading
 
 import pytest
+import requests
 
 from src import caldav_sync, caldav_writeback
 
@@ -26,12 +27,13 @@ def test_build_dav_client_disables_redirects():
 
 def test_dav_client_does_not_follow_redirect_to_internal_host():
     """End-to-end through the real DAVClient: a 302 toward an internal host
-    must NOT be followed. Without the fix the sink is contacted (SSRF); with it
+    must NOT be followed. Exercise the session limit separately because some
-    the redirect is refused and the sink is never reached."""
+    caldav versions independently refuse cross-origin PROPFIND redirects."""
     pytest.importorskip("caldav")
 
     sink_hits: list[str] = []
     public_methods: list[str] = []
+    same_origin_hits: list[str] = []
 
     class _Internal(http.server.BaseHTTPRequestHandler):
         # Stand-in for an internal service the attacker redirects toward.
@@ -48,9 +50,15 @@ def test_dav_client_does_not_follow_redirect_to_internal_host():
     class _Public(http.server.BaseHTTPRequestHandler):
         # The "validated" public CalDAV server that redirects everything inward.
         def do_GET(self):  # noqa: N802
+            if self.path == "/landing":
+                same_origin_hits.append(self.path)
+                self.send_response(200)
+                self.end_headers()
+                return
             public_methods.append(self.command)
             self.send_response(302)
-            self.send_header("Location", f"http://127.0.0.1:{internal_port}/leak")
+            location = "/landing" if self.path == "/same" else f"http://127.0.0.1:{internal_port}/leak"
+            self.send_header("Location", location)
             self.end_headers()
 
         do_PROPFIND = do_GET
@@ -62,8 +70,10 @@ def test_dav_client_does_not_follow_redirect_to_internal_host():
     internal_port = internal.server_address[1]
     public = socketserver.TCPServer(("127.0.0.1", 0), _Public)
     public_port = public.server_address[1]
-    threading.Thread(target=internal.serve_forever, daemon=True).start()
+    internal_thread = threading.Thread(target=internal.serve_forever, daemon=True)
-    threading.Thread(target=public.serve_forever, daemon=True).start()
+    public_thread = threading.Thread(target=public.serve_forever, daemon=True)
+    internal_thread.start()
+    public_thread.start()
     try:
         public_url = f"http://127.0.0.1:{public_port}/dav"
         client = caldav_sync._build_dav_client(public_url, "u", "p")
@@ -82,17 +92,35 @@ def test_dav_client_does_not_follow_redirect_to_internal_host():
         assert public_methods in (["PROPFIND"], ["PROPFIND", "GET"]), public_methods
         assert sink_hits == [], "redirect toward an internal host must not be followed"
 
-        # Positive control: demonstrate that the sink would see the redirect
+        # GET redirects are handled by the HTTP session even in caldav 3.7.
-        # if the session's redirect protection were removed.
+        # This makes the cross-port SSRF assertion sensitive to our own limit.
+        with pytest.raises(Exception, match="Exceeded 0 redirects"):
+            client.request(public_url, "GET", "")
+        assert sink_hits == [], "GET redirect toward the internal sink must be blocked"
+
+        # requests can reach the cross-port sink when redirects are enabled,
+        # even if DAVClient independently rejects cross-origin PROPFIND redirects.
+        with requests.Session() as session:
+            response = session.get(public_url, timeout=5)
+            assert response.status_code == 207
+        assert sink_hits == ["/leak"], "the cross-port redirect must reach the sink"
+
+        # GET follows redirects at the HTTP session layer across these caldav
+        # versions. Check the application's limit on a same-origin redirect too.
+        same_url = f"http://127.0.0.1:{public_port}/same"
+        with pytest.raises(Exception, match="Exceeded 0 redirects"):
+            client.request(same_url, "GET", "")
+        assert same_origin_hits == []
         client.session.max_redirects = 30
-        try:
+        client.request(same_url, "GET", "")
-            client.request(public_url, "PROPFIND", "")
+        assert same_origin_hits == ["/landing"]
-        except Exception:
-            pass  # The sink's response is intentionally not a valid DAV body.
-        assert sink_hits == ["/leak"], "the test must detect a followed redirect"
     finally:
         internal.shutdown()
         public.shutdown()
+        internal_thread.join()
+        public_thread.join()
+        internal.server_close()
+        public.server_close()
 
 
 def test_sync_and_writeback_construct_clients_through_the_helper():

tests/test_calendar_config_body.py

@@ -0,0 +1,35 @@
+"""The CalDAV config routes answer 400 for a body that is JSON but not an object."""
+import asyncio
+import json
+from types import SimpleNamespace
+
+import pytest
+from fastapi import HTTPException
+
+from routes.calendar_routes import _json_object_body
+
+
+def _request(raw: bytes):
+    async def _json():
+        return json.loads(raw)
+    return SimpleNamespace(json=_json)
+
+
+def _parse(raw: bytes):
+    return asyncio.run(_json_object_body(_request(raw)))
+
+
+def test_object_body_is_returned():
+    assert _parse(b'{"url": "https://dav.example/"}') == {"url": "https://dav.example/"}
+
+
+@pytest.mark.parametrize("raw", [b"", b"not json"])
+def test_unreadable_body_counts_as_empty(raw):
+    assert _parse(raw) == {}
+
+
+@pytest.mark.parametrize("raw", [b"[]", b'"x"', b"5", b"null"])
+def test_non_object_body_is_a_400(raw):
+    with pytest.raises(HTTPException) as exc:
+        _parse(raw)
+    assert exc.value.status_code == 400

tests/test_chat_tidy_content_preservation.py

@@ -0,0 +1,148 @@
+"""Both chat tidy entry points retain saved conversations and still remove empties."""
+import asyncio
+from datetime import datetime, timedelta, timezone
+
+import pytest
+from fastapi import APIRouter, FastAPI
+from fastapi.testclient import TestClient
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+
+import core.database as database
+import core.session_manager as manager_module
+import routes.session_routes as routes
+import src.session_actions as actions
+
+
+@pytest.fixture
+def tidy_api(tmp_path, monkeypatch):
+    engine = create_engine(
+        f"sqlite:///{(tmp_path / 'tidy.db').as_posix()}",
+        connect_args={"check_same_thread": False},
+    )
+    database.Base.metadata.create_all(engine)
+    factory = sessionmaker(bind=engine, autoflush=False)
+    monkeypatch.setattr(database, "SessionLocal", factory)
+    monkeypatch.setattr(manager_module, "SessionLocal", factory)
+    monkeypatch.setattr(routes, "SessionLocal", factory)
+    monkeypatch.setattr(routes, "router", APIRouter(prefix="/api"))
+    monkeypatch.setattr(routes, "effective_user", lambda request: "alice")
+    manager = manager_module.SessionManager.__new__(manager_module.SessionManager)
+    manager.sessions = {}
+    manager.upload_handler = None
+    monkeypatch.setattr(manager, "get_sessions_for_user", lambda owner: {})
+    app = FastAPI()
+    app.include_router(routes.setup_session_routes(manager, {}))
+    with TestClient(app) as client:
+        yield client, factory, manager
+    engine.dispose()
+
+
+def run_tidy(client, entrypoint):
+    if entrypoint == "action":
+        return asyncio.run(actions.run_auto_sort("alice", skip_llm=True))
+    response = client.post("/api/sessions/auto-sort?skip_llm=true")
+    assert response.status_code == 200, response.text
+    return response.json()
+
+
+@pytest.mark.parametrize("entrypoint", ["route", "action"])
+@pytest.mark.parametrize("title,prompt", [
+    ("New chat", "Please analyze the following contract and retain all details. " * 20),
+    ("Contract analysis", "Review contract"),
+    ("Chat: analysis", "Explain these results and retain the full answer"),
+    ("Test", "Here is a large input that should be analyzed and retained"),
+], ids=["generic-title", "short-prompt", "title-prefix", "test-title"])
+def test_tidy_preserves_nontrivial_saved_answer(tidy_api, title, prompt, entrypoint):
+    client, factory, manager = tidy_api
+    old = datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=2)
+    answer = "Clause analysis and recommended changes.\n" * 2000
+    with factory() as db:
+        db.add(database.Session(
+            id="valuable-chat", name=title, owner="alice", endpoint_url="", model="",
+            archived=False, is_important=False, message_count=2,
+            created_at=old, updated_at=old, last_accessed=old, last_message_at=old,
+        ))
+        db.commit()
+        db.add_all([
+            database.ChatMessage(id="prompt", session_id="valuable-chat", role="user",
+                                 content=prompt, timestamp=old),
+            database.ChatMessage(id="answer", session_id="valuable-chat", role="assistant",
+                                 content=answer, timestamp=old + timedelta(seconds=1)),
+        ])
+        db.commit()
+    run_tidy(client, entrypoint)
+    with factory() as db:
+        remaining = db.query(database.ChatMessage).filter_by(session_id="valuable-chat").count()
+        survives = db.get(database.Session, "valuable-chat") is not None
+    assert survives and remaining == 2, "Title/prompt heuristics must not erase saved answers"
+    with factory() as db:
+        assert db.get(database.ChatMessage, "answer").content == answer
+
+
+@pytest.mark.parametrize("entrypoint", ["route", "action"])
+@pytest.mark.parametrize("kind", ["unanswered", "greeting", "empty", "fresh-empty", "starred-empty", "archived", "other-owner", "incognito"])
+def test_tidy_cleanup_boundaries(tidy_api, entrypoint, kind):
+    client, factory, manager = tidy_api
+    now = datetime.now(timezone.utc).replace(tzinfo=None)
+    timestamp = now if kind == "fresh-empty" else now - timedelta(hours=2)
+    message = kind in {"unanswered", "greeting", "archived", "other-owner", "incognito"}
+    with factory() as db:
+        db.add(database.Session(
+            id="boundary-chat", name="Incognito" if kind == "incognito" else "New chat",
+            owner="bob" if kind == "other-owner" else "alice", endpoint_url="", model="",
+            archived=kind == "archived", is_important=kind == "starred-empty",
+            # Deliberately inaccurate metadata: cleanup must use message rows.
+            message_count=0 if message else 40,
+            created_at=timestamp, updated_at=timestamp, last_accessed=timestamp, last_message_at=timestamp,
+        ))
+        db.commit()
+        if message:
+            db.add(database.ChatMessage(id="boundary-message", session_id="boundary-chat", role="user",
+                                        content="hi" if kind == "greeting" else "Pending question " * 100,
+                                        timestamp=timestamp))
+            db.commit()
+    run_tidy(client, entrypoint)
+    deleted = kind in {"empty", "incognito"}
+    with factory() as db:
+        assert (db.get(database.Session, "boundary-chat") is None) == deleted
+        assert db.query(database.ChatMessage).filter_by(session_id="boundary-chat").count() == int(message and not deleted)
+
+
+@pytest.mark.parametrize("entrypoint", ["route", "action"])
+def test_tidy_keeps_folder_assignment_working(tidy_api, monkeypatch, entrypoint):
+    from types import SimpleNamespace
+    import src.llm_core as llm
+    import src.task_endpoint as endpoints
+    client, factory, manager = tidy_api
+    old = datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=2)
+    for sid in ["chat-one-long-id", "chat-two-long-id"]:
+        with factory() as db:
+            db.add(database.Session(id=sid, name="New chat", owner="alice", endpoint_url="", model="",
+                                    archived=False, created_at=old, updated_at=old,
+                                    last_accessed=old, last_message_at=old))
+            db.commit()
+            db.add(database.ChatMessage(id=f"{sid}-message", session_id=sid, role="user",
+                                        content="Saved question", timestamp=old))
+            db.commit()
+        manager.sessions[sid] = SimpleNamespace(id=sid, name="New chat", archived=False,
+                                                updated_at=old.isoformat(), created_at=old.isoformat())
+    monkeypatch.setattr(manager, "get_sessions_for_user", lambda owner: manager.sessions)
+    monkeypatch.setattr(endpoints, "resolve_task_endpoint", lambda *args, **kwargs: ("http://model.invalid", "test", {}))
+    response_text = '{"folders":{"Projects":["chat-one","chat-two"]}}'
+    monkeypatch.setattr(llm, "llm_call", lambda *args, **kwargs: response_text)
+
+    async def async_response(*args, **kwargs):
+        return response_text
+
+    monkeypatch.setattr(llm, "llm_call_async", async_response)
+    if entrypoint == "action":
+        result = asyncio.run(actions.run_auto_sort("alice"))
+        assert "Sorted 2 sessions" in result
+    else:
+        response = client.post("/api/sessions/auto-sort")
+        assert response.status_code == 200, response.text
+        assert response.json()["updated"] == 2
+    with factory() as db:
+        assert db.query(database.ChatMessage).count() == 2
+        assert {row.folder for row in db.query(database.Session).all()} == {"Projects"}

tests/test_checkin_digest_owner_scope.py

@@ -7,13 +7,13 @@ CalendarCal.owner; the query must join it, like routes/calendar_routes.
 """
 import uuid
 import sys
-from datetime import datetime
+from datetime import datetime, timedelta
 
 import pytest
 from tests.helpers.database import disposable_database
 
 from core.database import CalendarEvent, CalendarCal
-from src.task_scheduler import _checkin_calendar_events
+from src.task_scheduler import _checkin_calendar_events, _digest_windows
 
 
 @pytest.fixture(autouse=True)
@@ -68,3 +68,44 @@ def test_cancelled_excluded_and_window_respected():
         assert "cancelled" not in [ev.summary for ev in out]
     finally:
         db.close()
+
+
+@pytest.mark.parametrize("days, expected_label", [(2, "this_week"), (7, "next_30_days")])
+def test_boundary_event_appears_in_only_one_digest_window(days, expected_label):
+    now = datetime(2026, 6, 1, 9)
+    boundary = now + timedelta(days=days)
+    with _TS() as db:
+        db.add(CalendarCal(id="boundary-cal", owner="alice", name="Boundary"))
+        db.add(CalendarEvent(
+            uid="boundary-event", calendar_id="boundary-cal", summary="Boundary meeting",
+            dtstart=boundary, dtend=boundary + timedelta(hours=1), status="confirmed",
+        ))
+        db.commit()
+
+        buckets = [
+            label for label, start, end in _digest_windows(now)
+            if _checkin_calendar_events(db, "alice", start, end)
+        ]
+        assert buckets == [expected_label]
+
+
+def test_digest_includes_start_and_excludes_end():
+    start = datetime(2026, 6, 1, 9)
+    end = start + timedelta(days=2)
+    with _TS() as db:
+        db.add(CalendarCal(id="bounds-cal", owner="alice", name="Bounds"))
+        for uid, when in [
+            ("before", start - timedelta(seconds=1)),
+            ("start", start),
+            ("inside", end - timedelta(seconds=1)),
+            ("end", end),
+        ]:
+            db.add(CalendarEvent(
+                uid=uid, calendar_id="bounds-cal", summary=uid, dtstart=when,
+                dtend=when + timedelta(hours=1), status="confirmed",
+            ))
+        db.commit()
+
+        assert [ev.uid for ev in _checkin_calendar_events(db, "alice", start, end)] == [
+            "start", "inside",
+        ]

tests/test_compare_stream_contract_browser.py

@@ -0,0 +1,70 @@
+"""Compare's actual stream consumer must preserve evidence and terminal state."""
+import json
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+
+ROOT = Path(__file__).resolve().parents[1]
+
+CASES = [
+    ("lf", 'data: {"delta":"Hello"}\n\ndata: [DONE]\n\n', False, True, "Hello"),
+    ("crlf", 'data:{"delta":"Héllö"}\r\n\r\ndata:[DONE]\r\n\r\n', False, True, "Héllö"),
+    ("multiline", 'data: {"delta":\ndata: "Hello"}\n\ndata: [DONE]\n\n', False, True, "Hello"),
+    ("eof", 'data: {"delta":"Partial"}\n\n', False, False, "ended before completion"),
+    ("named-error", 'event: error\ndata: {"text":"Provider Z private-model failed"}\n\n', False, False, "Provider Z private-model failed"),
+    ("typed-error", 'data: {"type":"error","data":{"message":"Provider Z <img src=x onerror=alert(1)>"}}\n\ndata: [DONE]\n\n', False, False, "Provider Z <img"),
+    ("blind-error", 'event:error\ndata:{"message":"Provider Z private-model failed"}\n\n', True, False, "request failed"),
+    ("empty-error", 'event:error\ndata:\n\n', False, False, "request failed"),
+    ("done-open", 'data: {"delta":"Hello"}\n\ndata: [DONE]\n\n', False, True, "Hello"),
+    ("heartbeat", 'data: {"delta":"Hello"}\n\ndata: [DONE]\n\n', False, True, "Hello"),
+    ("cr", 'data: {"delta":"Hello"}\r\rdata:[DONE]\r\r', False, True, "Hello"),
+    ("invalid", 'data: not-json\n\ndata: [DONE]\n\n', False, False, "Invalid model stream data"),
+]
+
+
+@pytest.mark.parametrize("kind,wire,blind,success,text", CASES)
+def test_compare_stream_contract(kind, wire, blind, success, text):
+    if not shutil.which("node"):
+        pytest.skip("node not installed")
+    script = r"""
+const assert=require('node:assert/strict');
+let chromium;try{({chromium}=require('playwright'));}catch(e){if(e.code==='MODULE_NOT_FOUND')process.exit(77);throw e;}
+(async()=>{
+ const browser=await chromium.launch({headless:true});
+ try{
+  const page=await browser.newPage();
+  await page.goto(process.env.ODYSSEUS_TEST_STATIC_ORIGIN+'/static/js/storage.js');
+  await page.setContent('<div class="compare-pane" data-pane="0"><div id="cmp-badge-0"></div><div id="cmp-history-0"><div class="msg msg-ai"><div class="body"></div></div></div></div>');
+  const result=await page.evaluate(async(data)=>{
+   const {default:state}=await import('/static/js/compare/state.js');
+   const {streamToPane}=await import('/static/js/compare/stream.js');
+   Object.assign(state,{isActive:true,_selectedModels:[{model:'fixture',name:'Fixture'}],_paneSessionIds:['session'],_compareMode:'chat',_parallel:true,_blindMode:data.blind,_timeout:data.kind==='heartbeat'?1:5});
+   const nativeFetch=window.fetch,bytes=new TextEncoder().encode(data.wire);let index=0,cancelled=false,context=null;
+   window.fetch=async(url,options)=>{
+    if(!String(url).endsWith('/api/chat_stream'))return nativeFetch(url,options);
+    context=options.body.get('search_context');
+    const heartbeats=[...Array(6).fill(': ping\n\n'),data.wire];
+    return new Response(new ReadableStream({start(controller){options.signal.addEventListener('abort',()=>controller.error(new DOMException('Stopped','AbortError')),{once:true});},async pull(controller){if(data.kind==='heartbeat'){if(index<heartbeats.length){await new Promise(r=>setTimeout(r,200));controller.enqueue(new TextEncoder().encode(heartbeats[index++]));}else controller.close();}else if(index<bytes.length)controller.enqueue(bytes.slice(index,index+=1));else if(data.kind!=='done-open')controller.close();},cancel(){cancelled=true;}}),{headers:{'Content-Type':'text/event-stream'}});
+   };
+   const message=document.querySelector('.msg-ai'),spinner=document.createElement('span');let spinnerDestroyed=false;spinner.textContent='Loading';message.querySelector('.body').appendChild(spinner);message._spinner={element:spinner,destroy(){spinnerDestroyed=true;spinner.remove();}};
+   await Promise.race([streamToPane(0,'session','Prompt',message,{skipBadge:true}),new Promise((_,reject)=>setTimeout(()=>reject(Error('Consumer did not finish at DONE')),data.kind==='heartbeat'?6000:1500))]);
+   const pane=document.querySelector('.compare-pane');
+   return {done:pane.classList.contains('is-done'),failed:pane.classList.contains('is-failed'),text:pane.innerText,images:pane.querySelectorAll('img').length,cancelled,spinnerDestroyed,spinnerOwned:!!message._spinner};
+  },JSON.parse(process.argv[1]));
+  const expected=JSON.parse(process.argv[1]);
+  assert.equal(result.done,expected.success);assert.equal(result.failed,!expected.success);assert(result.text.includes(expected.text),JSON.stringify(result));assert.equal(result.images,0);
+  assert(result.spinnerDestroyed);assert(!result.spinnerOwned);
+  if(expected.blind)assert(!/Provider Z|private-model/.test(result.text),JSON.stringify(result));
+  if(expected.kind==='done-open')assert(result.cancelled);
+  console.log(JSON.stringify(result));
+ }finally{await browser.close();}
+})().catch(e=>{console.error(e);process.exitCode=1;});
+"""
+    data = dict(kind=kind, wire=wire, blind=blind, success=success, text=text)
+    result = subprocess.run(["node", "-e", script, json.dumps(data)], cwd=ROOT,
+                            capture_output=True, text=True, timeout=30)
+    if result.returncode == 77:
+        pytest.skip("playwright not installed")
+    assert result.returncode == 0, result.stdout + result.stderr

tests/test_cookbook_dependency_completion_regression.py

@@ -1,4 +1,14 @@
+import os
+import subprocess
 from pathlib import Path
+import pytest
+from core.platform_compat import find_bash, git_bash_path
+from routes.cookbook_helpers import (
+    ODYSSEUS_LLAMA_SHIM_MARKER,
+    _append_llama_server_shim_definition_lines,
+    _append_llama_server_shim_reconciliation_lines,
+    _append_llama_server_finalization_lines,
+)
 
 from tests.helpers.stylesheets import app_css
 
@@ -615,3 +625,730 @@ def test_local_dependency_probe_refreshes_user_site_visibility():
     # deps (realesrgan) probe as not-installed until a restart. See #4810.
     assert "if user_site and os.path.isdir(user_site):" in source
     assert "site.addsitedir(user_site)" in source
+
+
+def test_llama_runner_path_preserves_standard_user_bin_order():
+    """The runner PATH must preserve standard user $HOME/bin precedence while
+    relying on ownership-aware reconciliation to retire Odysseus shims when
+    native builds exist."""
+    source = _read("routes/cookbook_routes.py")
+
+    assert (
+        'export PATH="$HOME/.local/bin:$HOME/bin:$HOME/llama.cpp/build/bin:/opt/homebrew/bin:/usr/local/bin:$PATH"'
+        in source
+    )
+
+
+def _run_shell(tmp_path, lines, cwd=None, path=None, extra_env=None):
+    from core.platform_compat import find_bash, git_bash_path
+    bash = find_bash() or "bash"
+    env = dict(os.environ, HOME=git_bash_path(tmp_path))
+    if path is not None:
+        env["PATH"] = path
+    if extra_env:
+        env.update(extra_env)
+    return subprocess.run(
+        [bash, "-c", "\n".join(lines)],
+        cwd=str(cwd) if cwd else None,
+        env=env,
+        capture_output=True,
+        text=True,
+    )
+
+
+def _write_odysseus_shim(path: Path) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(
+        f"#!/usr/bin/env bash\n{ODYSSEUS_LLAMA_SHIM_MARKER}\nexec python3 -m llama_cpp.server \"$@\"\n",
+        encoding="utf-8",
+    )
+    path.chmod(0o755)
+
+
+def _write_mock_native(path: Path, succeeds: bool = True) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    if succeeds:
+        content = '#!/usr/bin/env bash\nif [ "$1" = "--version" ]; then echo "version: 0.1.0-dev"; exit 0; fi\necho "RUN_NATIVE: $@"\nexit 0\n'
+    else:
+        content = '#!/usr/bin/env bash\necho "error: missing libllama.so" >&2\nexit 127\n'
+    path.write_text(content, encoding="utf-8")
+    path.chmod(0o755)
+
+
+def _write_mock_python3(path: Path, succeeds: bool = True) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(f"#!/usr/bin/env bash\nexit {0 if succeeds else 1}\n", encoding="utf-8")
+    path.chmod(0o755)
+
+
+def test_odysseus_create_llama_shim_never_overwrites_user_regular_file(tmp_path):
+    """Creation boundary: never overwrite a non-executable user-owned regular file."""
+    user_file = tmp_path / "bin/llama-server"
+    user_file.parent.mkdir(parents=True, exist_ok=True)
+    original_content = "user configuration or binary data without marker"
+    user_file.write_text(original_content, encoding="utf-8")
+
+    lines = []
+    _append_llama_server_shim_definition_lines(lines)
+    lines.append('_odysseus_create_llama_shim')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode != 0, "Creation should return non-zero when target is a user file"
+    assert user_file.read_text(encoding="utf-8") == original_content, "User file must remain byte-for-byte unchanged"
+
+
+def test_odysseus_create_llama_shim_never_overwrites_user_symlink(tmp_path):
+    """Creation boundary: never follow or overwrite a user-owned symlink."""
+    bin_dir = tmp_path / "bin"
+    bin_dir.mkdir(parents=True, exist_ok=True)
+    target_file = tmp_path / "user_target.txt"
+    target_file.write_text("precious user target content", encoding="utf-8")
+    user_symlink = bin_dir / "llama-server"
+    user_symlink.symlink_to(target_file)
+
+    lines = []
+    _append_llama_server_shim_definition_lines(lines)
+    lines.append('_odysseus_create_llama_shim')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode != 0, "Creation should return non-zero when target is a user symlink"
+    assert user_symlink.is_symlink(), "Symlink must remain intact"
+    assert target_file.read_text(encoding="utf-8") == "precious user target content", "Target file must not be modified"
+
+
+def test_odysseus_create_llama_shim_never_overwrites_dangling_symlink(tmp_path):
+    """Creation boundary: never overwrite a dangling user-owned symlink."""
+    bin_dir = tmp_path / "bin"
+    bin_dir.mkdir(parents=True, exist_ok=True)
+    dangling_symlink = bin_dir / "llama-server"
+    dangling_symlink.symlink_to(tmp_path / "nonexistent_target")
+
+    lines = []
+    _append_llama_server_shim_definition_lines(lines)
+    lines.append('_odysseus_create_llama_shim')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode != 0, "Creation should return non-zero when target is a dangling symlink"
+    assert dangling_symlink.is_symlink(), "Dangling symlink must remain intact"
+
+
+def test_odysseus_create_llama_shim_safely_replaces_existing_odysseus_shim(tmp_path):
+    """Creation boundary: safely replace an existing Odysseus-owned compatibility shim."""
+    shim_file = tmp_path / "bin/llama-server"
+    shim_file.parent.mkdir(parents=True, exist_ok=True)
+    shim_file.write_text(f"#!/bin/bash\n{ODYSSEUS_LLAMA_SHIM_MARKER}\n# old shim body\n", encoding="utf-8")
+
+    lines = []
+    _append_llama_server_shim_definition_lines(lines)
+    lines.append('_odysseus_create_llama_shim')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    new_content = shim_file.read_text(encoding="utf-8")
+    assert ODYSSEUS_LLAMA_SHIM_MARKER in new_content
+    assert "exec python3 -m llama_cpp.server" in new_content
+
+
+def test_odysseus_create_llama_shim_secure_temp_prevents_symlink_attack(tmp_path):
+    """Creation boundary: secure mktemp prevents predictable $$ temp symlink attacks.
+
+    Given:
+    - A victim file at ~/victim containing DO_NOT_TOUCH
+    - A pre-created symlink at ~/bin/.llama-server.odysseus.tmp.$$ pointing to ~/victim
+    - _odysseus_create_llama_shim executed within that exact same bash PID ($$)
+
+    Prove:
+    - ~/victim remains unmodified
+    - Pre-existing crafted symlink is NOT followed or overwritten
+    - Legitimate shim is successfully created at ~/bin/llama-server via secure temp path
+    """
+    lines = [
+        'victim="$HOME/victim"',
+        'printf "DO_NOT_TOUCH\\n" > "$victim"',
+        'mkdir -p "$HOME/bin"',
+        'old_tmp="$HOME/bin/.llama-server.odysseus.tmp.$$";',
+        'ln -s "$victim" "$old_tmp"',
+    ]
+    _append_llama_server_shim_definition_lines(lines)
+    lines.append('_odysseus_create_llama_shim')
+    lines.append('res=$?')
+    lines.append('echo "CREATION_RES=$res"')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert "CREATION_RES=0" in res.stdout
+
+    victim_file = tmp_path / "victim"
+    assert victim_file.exists()
+    assert victim_file.read_text(encoding="utf-8") == "DO_NOT_TOUCH\n", "Victim file must remain untouched"
+
+    target_shim = tmp_path / "bin/llama-server"
+    assert target_shim.exists()
+    assert ODYSSEUS_LLAMA_SHIM_MARKER in target_shim.read_text(encoding="utf-8")
+    assert target_shim.stat().st_mode & 0o111 != 0, "Shim must be executable"
+
+
+def test_stale_odysseus_shim_removed_when_python_server_unusable(tmp_path):
+    """Scenario 1: Stale Odysseus shim + missing Python server -> owned shim removed."""
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+
+    lines = []
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert not shim_file.exists(), "Stale Odysseus shim must be removed when python server is unavailable"
+
+
+def test_healthy_odysseus_shim_retained_when_no_native_server(tmp_path):
+    """Scenario 2: Healthy Odysseus shim + no native server -> shim preserved."""
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=True)
+
+    lines = []
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    res = _run_shell(tmp_path, lines, path=f"{git_bash_path(mock_py_dir)}:{os.environ.get('PATH', '')}")
+    assert res.returncode == 0
+    assert shim_file.exists(), "Healthy Odysseus shim must be preserved when no native server is present"
+
+
+def test_reconciliation_does_not_retire_healthy_shim_for_broken_native_binary(tmp_path):
+    """Reconciliation health check: a broken native candidate executable must NOT retire
+    a healthy Odysseus Python shim."""
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+    _write_mock_native(tmp_path / "llama.cpp/build/bin/llama-server", succeeds=False)
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=True)
+
+    lines = [f'export PATH="$HOME/llama.cpp/build/bin:{git_bash_path(mock_py_dir)}:$PATH"']
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert shim_file.exists(), "Healthy Odysseus shim must NOT be retired when native binary is broken"
+
+
+def test_odysseus_shim_retired_when_native_server_healthy(tmp_path):
+    """Scenario 3: Odysseus shim + healthy native server available -> native server wins."""
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+    _write_mock_native(tmp_path / "llama.cpp/build/bin/llama-server", succeeds=True)
+
+    lines = ['export PATH="$HOME/bin:$HOME/llama.cpp/build/bin:$PATH"']
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert not shim_file.exists(), "Odysseus shim must be retired when healthy native server is available"
+
+
+def test_user_owned_bin_llama_server_never_touched(tmp_path):
+    """Scenario 4: Arbitrary user ~/bin/llama-server -> untouched and not deprioritized."""
+    user_bin = tmp_path / "bin/llama-server"
+    user_bin.parent.mkdir(parents=True, exist_ok=True)
+    user_bin.write_text("#!/bin/bash\n# Custom user build script\nexec /opt/custom/llama-server\n", encoding="utf-8")
+    user_bin.chmod(0o755)
+
+    _write_mock_native(tmp_path / "llama.cpp/build/bin/llama-server", succeeds=True)
+
+    lines = ['export PATH="$HOME/bin:$HOME/llama.cpp/build/bin:$PATH"']
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert user_bin.exists(), "User-owned ~/bin/llama-server must never be deleted"
+    assert "Custom user build script" in user_bin.read_text(encoding="utf-8")
+
+
+def test_reconciliation_ignores_native_binary_outside_effective_path(tmp_path):
+    """Native candidate outside effective PATH must NOT be considered."""
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+    # Native binary placed in /usr_custom/bin which is NOT in PATH
+    _write_mock_native(tmp_path / "usr_custom/bin/llama-server", succeeds=True)
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=True)
+
+    lines = []
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    # PATH only contains ~/bin and mock_py_dir — usr_custom is omitted
+    path = f"{git_bash_path(tmp_path / 'bin')}:{git_bash_path(mock_py_dir)}"
+    res = _run_shell(tmp_path, lines, path=path)
+    assert res.returncode == 0
+    assert shim_file.exists(), "Shim must NOT be retired by native binary outside effective PATH"
+
+
+def test_reconciliation_skips_home_bin_with_or_without_trailing_slash(tmp_path):
+    """Reconciliation must exclude $HOME/bin and $HOME/bin/ from native candidate discovery."""
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=True)
+
+    lines = []
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    # PATH has ~/bin/ with trailing slash
+    path = f"{git_bash_path(tmp_path / 'bin')}/:{git_bash_path(mock_py_dir)}"
+    res = _run_shell(tmp_path, lines, path=path)
+    assert res.returncode == 0
+    assert shim_file.exists(), "Shim must NOT rediscover itself when PATH has trailing slash in $HOME/bin/"
+
+
+def test_reconciliation_does_not_discover_cwd_executable_when_path_omits_dot(tmp_path):
+    """Reconciliation must NOT synthesize an empty/trailing PATH element that searches cwd.
+
+    Given:
+    - Current working directory contains an executable `llama-server`
+    - Healthy Odysseus Python shim in ~/bin/llama-server
+    - PATH does NOT contain '.' or cwd
+    - No other healthy native binary on PATH
+
+    Prove:
+    - Reconciliation does not discover ./llama-server
+    - The healthy Odysseus shim in ~/bin/llama-server is preserved
+    """
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=True)
+
+    cwd_dir = tmp_path / "workdir"
+    _write_mock_native(cwd_dir / "llama-server", succeeds=True)
+
+    lines = []
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    path = f"{git_bash_path(tmp_path / 'bin')}:{git_bash_path(mock_py_dir)}:/usr/bin:/bin"
+    res = _run_shell(tmp_path, lines, cwd=cwd_dir, path=path)
+    assert res.returncode == 0
+    assert shim_file.exists(), "Healthy Odysseus shim must NOT be retired by finding ./llama-server in cwd"
+
+
+def test_reconciliation_discovers_healthy_native_on_explicit_other_path_entry(tmp_path):
+    """Reconciliation must discover healthy native binaries placed on explicit PATH entries."""
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+    custom_opt_bin = tmp_path / "custom_opt/bin"
+    _write_mock_native(custom_opt_bin / "llama-server", succeeds=True)
+
+    lines = []
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    path = f"{git_bash_path(tmp_path / 'bin')}:{git_bash_path(custom_opt_bin)}:/usr/bin:/bin"
+    res = _run_shell(tmp_path, lines, path=path)
+    assert res.returncode == 0
+    assert not shim_file.exists(), "Odysseus shim must be retired when healthy native server is on explicit PATH entry"
+
+
+def test_reconciliation_safely_iterates_path_with_glob_metacharacters(tmp_path):
+    """Reconciliation must not perform pathname expansion / globbing on PATH components.
+
+    Given:
+    - A directory path containing shell glob characters: ~/custom[dir]/bin
+    - A native binary ~/custom[dir]/bin/llama-server
+    - A working directory containing matching glob entries
+    - An existing Odysseus shim in ~/bin/llama-server
+
+    Prove:
+    - Reconciliation correctly parses the literal path without expanding [dir]
+    - Discovers the healthy native server and retires the Odysseus shim
+    """
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+
+    glob_dir = tmp_path / "custom[dir]/bin"
+    _write_mock_native(glob_dir / "llama-server", succeeds=True)
+
+    cwd_dir = tmp_path / "cwd"
+    cwd_dir.mkdir(parents=True, exist_ok=True)
+    (cwd_dir / "customd").mkdir(parents=True, exist_ok=True)
+    (cwd_dir / "customi").mkdir(parents=True, exist_ok=True)
+
+    lines = []
+    _append_llama_server_shim_reconciliation_lines(lines)
+
+    path = f"{git_bash_path(tmp_path / 'bin')}:{git_bash_path(glob_dir)}:/usr/bin:/bin"
+    res = _run_shell(tmp_path, lines, cwd=cwd_dir, path=path)
+    assert res.returncode == 0
+    assert not shim_file.exists(), "Reconciliation must discover literal bracketed PATH without globbing"
+
+
+def test_fallback_install_success_creates_usable_launcher(tmp_path):
+    """Scenario 5: No native + no Python server -> fallback install succeeds
+    -> usable compatibility launcher exists and resolves command -v llama-server."""
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=True)
+
+    lines = [f'export PATH="$HOME/bin:{git_bash_path(mock_py_dir)}:$PATH"']
+    _append_llama_server_shim_definition_lines(lines)
+    _append_llama_server_finalization_lines(lines)
+    lines.append('command -v llama-server')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    created_shim = tmp_path / "bin/llama-server"
+    assert created_shim.exists()
+    assert ODYSSEUS_LLAMA_SHIM_MARKER in created_shim.read_text(encoding="utf-8")
+    assert str(tmp_path / "bin/llama-server") in res.stdout
+
+
+def test_failed_fallback_sets_preflight_exit_127(tmp_path):
+    """Scenario 6: No native + failed fallback installation -> ODYSSEUS_PREFLIGHT_EXIT=127."""
+    lines = [
+        'ODYSSEUS_PREFLIGHT_EXIT=""',
+        'export PATH="$HOME/bin:$PATH"',
+    ]
+    _append_llama_server_shim_definition_lines(lines)
+    _append_llama_server_finalization_lines(lines)
+    lines.append('echo "EXIT=$ODYSSEUS_PREFLIGHT_EXIT"')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert "EXIT=127" in res.stdout
+
+
+def test_python_server_readiness_requires_llama_cpp_server_module():
+    """Scenario 7 & 8: Python server readiness requires `import llama_cpp.server`,
+    while GPU offload capability checks still use base `import llama_cpp`."""
+    routes_source = _read("routes/cookbook_routes.py")
+    helpers_source = _read("routes/cookbook_helpers.py")
+
+    # Windows fallback readiness check requires .server
+    assert 'try { python -c "import llama_cpp.server" 2>$null } catch {}' in routes_source
+    # Termux fallback readiness check requires .server
+    assert 'if ! python3 -c "import llama_cpp.server" 2>/dev/null; then' in routes_source
+    # Build-failure fallback pip check requires .server
+    assert 'if [ -z "$_odysseus_llama_bin" ] && ! python3 -c "import llama_cpp.server" 2>/dev/null; then' in routes_source
+    # Shim reconciliation and post-install finalization require .server
+    assert 'elif python3 -c "import llama_cpp.server" 2>/dev/null; then' in helpers_source
+    assert 'if [ -z "$_odysseus_llama_bin" ] && python3 -c "import llama_cpp.server" 2>/dev/null; then' in helpers_source
+
+    # GPU offload support checks still intentionally inspect base llama_cpp
+    assert 'llama_cpp.llama_supports_gpu_offload()' in routes_source
+
+
+def test_darwin_native_build_does_not_force_write_bin_symlink():
+    """Darwin source build must not overwrite ~/bin/llama-server with a forced symlink."""
+    source = _read("routes/cookbook_routes.py")
+
+    assert "&& ln -sf ~/llama.cpp/build/bin/llama-server ~/bin/llama-server" not in source
+    assert 'cmake --build build -j"$NPROC" --target llama-server' in source
+
+
+@pytest.mark.asyncio
+async def test_generated_runner_preflight_retires_stale_shim_before_backend_selection(monkeypatch, tmp_path):
+    """Integration regression: The actual generated Cookbook runner must execute
+    ownership reconciliation BEFORE evaluating the backend selection (`if Termux ... elif ! command -v llama-server`).
+
+    Given:
+    - Executable stale Odysseus shim in ~/bin/llama-server (with canonical marker)
+    - Healthy native llama-server in ~/llama.cpp/build/bin/llama-server
+    - ~/bin before ~/llama.cpp/build/bin on PATH
+
+    Prove:
+    1. Generated runner contains reconciliation shell lines strictly before the backend branching
+    2. Executing the generated runner script retires the stale ~/bin/llama-server shim
+    3. Resolves to the healthy native binary without attempting source build or pip fallback.
+    """
+    from unittest.mock import AsyncMock, MagicMock
+    from routes.cookbook_helpers import ServeRequest
+    import routes.cookbook_routes as cookbook_routes
+    from starlette.requests import Request
+
+    fake_home = tmp_path / "fake_home"
+    stale_shim = fake_home / "bin/llama-server"
+    _write_odysseus_shim(stale_shim)
+    native_bin = fake_home / "llama.cpp/build/bin/llama-server"
+    _write_mock_native(native_bin, succeeds=True)
+
+    # Mock environment to generate the real runner file with isolated DB
+    tmux_log_dir = tmp_path / "tmux_logs"
+    tmux_log_dir.mkdir(parents=True)
+    monkeypatch.setattr(cookbook_routes, "TMUX_LOG_DIR", tmux_log_dir)
+    monkeypatch.setattr(cookbook_routes, "require_admin", lambda request: None)
+    monkeypatch.setattr(cookbook_routes, "_binary_available", AsyncMock(return_value=True))
+    monkeypatch.setattr(cookbook_routes, "load_stored_hf_token", lambda **kwargs: "")
+    monkeypatch.setattr("core.database.SessionLocal", MagicMock())
+
+    class _Proc:
+        returncode = 0
+        stderr = None
+        async def wait(self):
+            return 0
+    monkeypatch.setattr(cookbook_routes.asyncio, "create_subprocess_shell", AsyncMock(return_value=_Proc()))
+
+    router = cookbook_routes.setup_cookbook_routes()
+    endpoint = None
+    for route in router.routes:
+        if route.path == "/api/model/serve" and "POST" in route.methods:
+            endpoint = route.endpoint
+            break
+    assert endpoint is not None
+
+    from types import SimpleNamespace
+    http_req = Request({
+        "type": "http",
+        "method": "POST",
+        "path": "/api/model/serve",
+        "headers": [],
+        "app": SimpleNamespace(state=SimpleNamespace(auth_manager=SimpleNamespace(
+            is_configured=True, is_admin=lambda user: user == "admin"))),
+    })
+    http_req.state.current_user = "admin"
+    http_req.state.local_model_authority = True
+
+    req = ServeRequest(
+        repo_id="test/repo",
+        cmd="llama-server -m /dummy/model.gguf --port 8085",
+        remote_host=None,
+    )
+
+    resp = await endpoint(http_req, req)
+    assert resp["ok"] is True
+    session_id = resp["session_id"]
+    runner_file = tmux_log_dir / f"{session_id}_run.sh"
+    assert runner_file.exists()
+    runner_script = runner_file.read_text(encoding="utf-8")
+
+    # Assert structural ordering in the generated script:
+    # Reconciliation must appear strictly BEFORE the Termux/backend check
+    recon_idx = runner_script.find("Native llama-server detected — retiring Odysseus compatibility shim in ~/bin")
+    termux_idx = runner_script.find("if [ -d /data/data/com.termux ]; then")
+    backend_idx = runner_script.find('elif [ -z "$_odysseus_llama_bin" ]; then')
+    assert recon_idx != -1, "Reconciliation block missing from generated runner"
+    assert termux_idx != -1, "Termux check missing from generated runner"
+    assert backend_idx != -1, "Backend selection missing from generated runner"
+    assert recon_idx < termux_idx < backend_idx, "Reconciliation must execute strictly before backend selection"
+
+    # Execute the generated runner preflight in a real subshell
+    res = _run_shell(fake_home, [str(runner_file)], path=f"{git_bash_path(fake_home / 'bin')}:{git_bash_path(fake_home / 'llama.cpp/build/bin')}:/usr/bin:/bin")
+    assert res.returncode == 0
+    assert not stale_shim.exists(), "Stale ~/bin/llama-server must have been retired by the generated preflight"
+    assert "RUN_NATIVE: -m /dummy/model.gguf --port 8085" in res.stdout
+    assert "Native llama-server not found — building from source" not in res.stdout
+
+
+def test_broken_first_native_candidate_healthy_second_candidate_executes_healthy(tmp_path):
+    """Regression finding P2: When PATH contains an earlier broken candidate followed by a
+    healthy candidate, reconciliation and launch must select and execute the healthy candidate."""
+    from core.platform_compat import git_bash_path
+    broken_dir = tmp_path / "broken_bin"
+    _write_mock_native(broken_dir / "llama-server", succeeds=False)
+
+    healthy_dir = tmp_path / "healthy_bin"
+    _write_mock_native(healthy_dir / "llama-server", succeeds=True)
+
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+
+    lines = [
+        f'export PATH="$HOME/bin:{git_bash_path(broken_dir)}:{git_bash_path(healthy_dir)}:/usr/bin:/bin"',
+    ]
+    _append_llama_server_shim_reconciliation_lines(lines)
+    lines.append('echo "RESOLVED=$_odysseus_llama_bin"')
+    lines.append('command -v llama-server')
+    lines.append('llama-server -m /dummy/model.gguf --port 8080')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert not shim_file.exists(), "Stale Odysseus shim must be retired"
+    assert f"RESOLVED={git_bash_path(healthy_dir / 'llama-server')}" in res.stdout
+    assert str(healthy_dir / "llama-server") in res.stdout
+    assert "RUN_NATIVE: -m /dummy/model.gguf --port 8080" in res.stdout
+    assert "error: missing libllama.so" not in res.stdout
+
+
+def test_paths_with_spaces_broken_first_healthy_second_launch(tmp_path):
+    """Regression: Paths containing spaces must be safely resolved, prioritized and launched."""
+    from core.platform_compat import git_bash_path
+    broken_dir = tmp_path / "broken path with spaces/bin"
+    _write_mock_native(broken_dir / "llama-server", succeeds=False)
+
+    healthy_dir = tmp_path / "healthy path with spaces/bin"
+    _write_mock_native(healthy_dir / "llama-server", succeeds=True)
+
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+
+    model_dir = tmp_path / "models with spaces"
+    model_dir.mkdir(parents=True, exist_ok=True)
+    model_file = model_dir / "my model.gguf"
+    model_file.write_text("dummy", encoding="utf-8")
+
+    lines = [
+        f'export PATH="$HOME/bin:{git_bash_path(broken_dir)}:{git_bash_path(healthy_dir)}:/usr/bin:/bin"',
+    ]
+    _append_llama_server_shim_reconciliation_lines(lines)
+    lines.append(f'llama-server -m "{git_bash_path(model_file)}" --port 8080')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert not shim_file.exists(), "Stale shim must be retired"
+    assert f"RUN_NATIVE: -m {git_bash_path(model_file)} --port 8080" in res.stdout
+    assert "error: missing libllama.so" not in res.stdout
+
+
+def test_command_hash_invalidation_switches_from_cached_broken_to_healthy(tmp_path):
+    """Regression: Command cache invalidation (`hash -r`) must prevent bash from reusing a
+    previously cached lookup to an earlier broken executable."""
+    from core.platform_compat import git_bash_path
+    broken_dir = tmp_path / "broken_bin"
+    _write_mock_native(broken_dir / "llama-server", succeeds=False)
+
+    healthy_dir = tmp_path / "healthy_bin"
+    _write_mock_native(healthy_dir / "llama-server", succeeds=True)
+
+    lines = [
+        # Populate bash hash cache with broken binary
+        f'export PATH="{git_bash_path(broken_dir)}:/usr/bin:/bin"',
+        'llama-server >/dev/null 2>&1 || true',
+        'hash',
+        # Now add healthy candidate later in PATH and run reconciliation
+        f'export PATH="$HOME/bin:{git_bash_path(broken_dir)}:{git_bash_path(healthy_dir)}:/usr/bin:/bin"',
+    ]
+    _append_llama_server_shim_reconciliation_lines(lines)
+    lines.append('llama-server -m /dummy/test.gguf')
+    lines.append('hash')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert "RUN_NATIVE: -m /dummy/test.gguf" in res.stdout
+    # Verify hash table ended with healthy binary, not broken
+    assert str(healthy_dir / "llama-server") in res.stdout
+
+
+def test_broken_native_candidate_falls_back_to_healthy_python_compatibility_shim(tmp_path):
+    """Regression: A broken native candidate must not shadow or retire a healthy Python shim."""
+    from core.platform_compat import git_bash_path
+    broken_dir = tmp_path / "broken_bin"
+    _write_mock_native(broken_dir / "llama-server", succeeds=False)
+
+    shim_file = tmp_path / "bin/llama-server"
+    _write_odysseus_shim(shim_file)
+
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=True)
+
+    lines = [
+        f'export PATH="$HOME/bin:{git_bash_path(broken_dir)}:{git_bash_path(mock_py_dir)}:/usr/bin:/bin"',
+    ]
+    _append_llama_server_shim_reconciliation_lines(lines)
+    lines.append('echo "RESOLVED=$_odysseus_llama_bin"')
+    lines.append('command -v llama-server')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert shim_file.exists(), "Healthy Python compatibility shim must be preserved"
+    assert f"RESOLVED={git_bash_path(shim_file)}" in res.stdout
+    assert str(shim_file) in res.stdout
+
+
+def test_broken_native_and_missing_python_backend_sets_preflight_127(tmp_path):
+    """Regression: When the only native candidates are broken and Python server is unavailable,
+    finalization must fail closed with ODYSSEUS_PREFLIGHT_EXIT=127 rather than claiming readiness."""
+    from core.platform_compat import git_bash_path
+    broken_dir = tmp_path / "broken_bin"
+    _write_mock_native(broken_dir / "llama-server", succeeds=False)
+
+    mock_py_dir = tmp_path / "mock_bin"
+    _write_mock_python3(mock_py_dir / "python3", succeeds=False)
+
+    lines = [
+        'ODYSSEUS_PREFLIGHT_EXIT=""',
+        f'export PATH="$HOME/bin:{git_bash_path(broken_dir)}:{git_bash_path(mock_py_dir)}:/usr/bin:/bin"',
+    ]
+    _append_llama_server_shim_definition_lines(lines)
+    _append_llama_server_shim_reconciliation_lines(lines)
+    _append_llama_server_finalization_lines(lines)
+    lines.append('echo "EXIT=$ODYSSEUS_PREFLIGHT_EXIT"')
+
+    res = _run_shell(tmp_path, lines)
+    assert res.returncode == 0
+    assert "EXIT=127" in res.stdout
+
+
+@pytest.mark.asyncio
+async def test_generated_runner_broken_first_healthy_second_e2e(monkeypatch, tmp_path):
+    """Integration regression: The full generated Cookbook runner script must launch the healthy
+    native candidate even when an earlier broken candidate exists on PATH."""
+    from unittest.mock import AsyncMock, MagicMock
+    from types import SimpleNamespace
+    from routes.cookbook_helpers import ServeRequest
+    import routes.cookbook_routes as cookbook_routes
+    from starlette.requests import Request
+    from core.platform_compat import git_bash_path
+
+    fake_home = tmp_path / "fake_home"
+    stale_shim = fake_home / "bin/llama-server"
+    _write_odysseus_shim(stale_shim)
+
+    broken_bin = fake_home / "broken/bin/llama-server"
+    _write_mock_native(broken_bin, succeeds=False)
+
+    healthy_bin = fake_home / "healthy/bin/llama-server"
+    _write_mock_native(healthy_bin, succeeds=True)
+
+    tmux_log_dir = tmp_path / "tmux_logs"
+    tmux_log_dir.mkdir(parents=True)
+    monkeypatch.setattr(cookbook_routes, "TMUX_LOG_DIR", tmux_log_dir)
+    monkeypatch.setattr(cookbook_routes, "require_admin", lambda request: None)
+    monkeypatch.setattr(cookbook_routes, "_binary_available", AsyncMock(return_value=True))
+    monkeypatch.setattr(cookbook_routes, "load_stored_hf_token", lambda **kwargs: "")
+    monkeypatch.setattr("core.database.SessionLocal", MagicMock())
+
+    class _Proc:
+        returncode = 0
+        stderr = None
+        async def wait(self):
+            return 0
+    monkeypatch.setattr(cookbook_routes.asyncio, "create_subprocess_shell", AsyncMock(return_value=_Proc()))
+
+    router = cookbook_routes.setup_cookbook_routes()
+    endpoint = None
+    for route in router.routes:
+        if route.path == "/api/model/serve" and "POST" in route.methods:
+            endpoint = route.endpoint
+            break
+    assert endpoint is not None
+
+    http_req = Request({
+        "type": "http",
+        "method": "POST",
+        "path": "/api/model/serve",
+        "headers": [],
+        "app": SimpleNamespace(state=SimpleNamespace(auth_manager=SimpleNamespace(
+            is_configured=True, is_admin=lambda user: user == "admin"))),
+    })
+    http_req.state.current_user = "admin"
+    http_req.state.local_model_authority = True
+
+    req = ServeRequest(
+        repo_id="test/repo",
+        cmd="llama-server -m /dummy/model.gguf --port 8090",
+        remote_host=None,
+    )
+
+    resp = await endpoint(http_req, req)
+    assert resp["ok"] is True
+    session_id = resp["session_id"]
+    runner_file = tmux_log_dir / f"{session_id}_run.sh"
+    assert runner_file.exists()
+
+    # Broken is earlier on PATH than healthy
+    res = _run_shell(
+        fake_home,
+        [str(runner_file)],
+        path=f"{git_bash_path(fake_home / 'bin')}:{git_bash_path(broken_bin.parent)}:{git_bash_path(healthy_bin.parent)}:/usr/bin:/bin",
+    )
+    assert res.returncode == 0
+    assert not stale_shim.exists(), "Stale shim must have been retired"
+    assert "RUN_NATIVE: -m /dummy/model.gguf --port 8090" in res.stdout
+    assert "error: missing libllama.so" not in res.stdout
+    assert "Native llama-server not found — building from source" not in res.stdout

tests/test_cookbook_recipe_panel_toggle.py

@@ -0,0 +1,38 @@
+"""Guard the only entry point to the Cookbook → Dependencies recipe panel.
+
+The recipe panel (model picker, Pip/Docker toggle, commands and Run) is
+rendered hidden, and only the `[data-dep-recipe-toggle]` caret shows it.
+The HW Fit and serve-card deep links (`openCookbookDependencies(..., {
+expandRecipe })`) open it by clicking that same caret. When the caret was
+dropped from the row markup, nothing could open the panel (#6747).
+"""
+import re
+from pathlib import Path
+
+JS_DIR = Path(__file__).resolve().parent.parent / "static" / "js"
+COOKBOOK_JS = JS_DIR / "cookbook.js"
+DIAGNOSIS_JS = JS_DIR / "cookbook-diagnosis.js"
+
+
+def _row_renderer(source: str) -> str:
+    start = source.index("const hasRecipe = RECIPE_BACKENDS.has(pkg.name);")
+    end = source.index("+ recipePanel;", start)
+    return source[start:end]
+
+
+def test_recipe_rows_render_the_panel_toggle():
+    row = _row_renderer(COOKBOOK_JS.read_text(encoding="utf-8"))
+
+    caret = re.search(r"const recipeCaret = hasRecipe\s*\?\s*`([^`]*)`\s*:\s*'';", row)
+    assert caret, "recipe rows must render a caret, or the recipe panel cannot be opened"
+    assert 'data-dep-recipe-toggle="${esc(pkg.name)}"' in caret.group(1)
+    assert 'aria-expanded="false"' in caret.group(1)
+    assert "+ recipeCaret" in row
+
+
+def test_toggle_handler_and_deep_link_use_the_rendered_attribute():
+    cookbook = COOKBOOK_JS.read_text(encoding="utf-8")
+    diagnosis = DIAGNOSIS_JS.read_text(encoding="utf-8")
+
+    assert "list.querySelectorAll('[data-dep-recipe-toggle]')" in cookbook
+    assert "row.querySelector('[data-dep-recipe-toggle]')" in diagnosis

tests/test_cookbook_stale_cache_markers.py

@@ -0,0 +1,83 @@
+"""A completed blob's empty retry marker must not keep its model stalled."""
+import json
+import os
+import subprocess
+import sys
+
+import pytest
+
+from routes.cookbook_helpers import _cached_model_scan_script
+from routes.cookbook_output import HF_CACHE_COMPLETE_PROBE, HF_CACHE_INCOMPLETE_PROBE
+
+_REPO = "fixture/stale-marker"
+_STATES = ["complete", "stale", "empty-unpaired", "partial", "partial-paired", "empty-paired-empty"]
+
+
+@pytest.fixture(params=["custom", "hf-home", "hub-env"])
+def cache(request, tmp_path):
+    root = tmp_path / "cache with spaces"
+    model = root / "hub" / ("models--" + _REPO.replace("/", "--"))
+    blobs = model / "blobs"
+    blobs.mkdir(parents=True)
+    snapshot = model / "snapshots" / "revision"
+    snapshot.mkdir(parents=True)
+    (snapshot / "config.json").write_text("{}")
+    env = {key: value for key, value in os.environ.items()
+           if key not in {"HF_HOME", "HUGGINGFACE_HUB_CACHE", "HF_HUB_CACHE"}}
+    if request.param == "custom":
+        argument = str(root)
+        scan = _cached_model_scan_script(add_hf_cache=str(root / "hub"))
+    else:
+        argument = ""
+        env["HF_HOME" if request.param == "hf-home" else "HUGGINGFACE_HUB_CACHE"] = str(
+            root if request.param == "hf-home" else root / "hub"
+        )
+        scan = _cached_model_scan_script()
+    return blobs, argument, env, scan
+
+
+def _seed(blobs, state):
+    if state in {"complete", "stale", "partial-paired"}:
+        (blobs / "sha").write_bytes(b"completed model data")
+    elif state == "empty-paired-empty":
+        (blobs / "sha").touch()
+    if state != "complete":
+        (blobs / "sha.incomplete").write_bytes(b"partial" if state.startswith("partial") else b"")
+
+
+@pytest.mark.parametrize("state", _STATES)
+@pytest.mark.parametrize("probe", [HF_CACHE_COMPLETE_PROBE, HF_CACHE_INCOMPLETE_PROBE], ids=["complete", "incomplete"])
+def test_cache_status_probes_distinguish_stale_and_partial(cache, state, probe):
+    blobs, argument, env, _ = cache
+    _seed(blobs, state)
+    before = {path.name: path.read_bytes() for path in blobs.iterdir()}
+    result = subprocess.run([sys.executable, "-c", probe, _REPO, argument],
+                            env=env, capture_output=True, text=True, timeout=15)
+    complete = state in {"complete", "stale"}
+    expected = complete if probe == HF_CACHE_COMPLETE_PROBE else not complete
+    assert result.returncode == (0 if expected else 1), result.stderr
+    assert {path.name: path.read_bytes() for path in blobs.iterdir()} == before
+
+
+@pytest.mark.parametrize("state", _STATES)
+def test_generated_inventory_agrees_without_mutating_cache(cache, state):
+    blobs, _, env, scan = cache
+    _seed(blobs, state)
+    before = {path.name: path.read_bytes() for path in blobs.iterdir()}
+    # Inventory's unrelated Ollama discovery stays off the network. Execute
+    # the actual generated scanner, including its cache-root resolution.
+    harness = (
+        "import urllib.request,shutil\n"
+        "def offline(*args,**kwargs): raise OSError('Ollama outside this fixture')\n"
+        "urllib.request.urlopen=offline\nshutil.which=lambda name: None\n"
+        + scan
+    )
+    result = subprocess.run([sys.executable, "-c", harness], env=env,
+                            capture_output=True, text=True, timeout=15)
+    assert result.returncode == 0, result.stderr
+    row = next(item for item in json.loads(result.stdout) if item["repo_id"] == _REPO)
+    assert row["has_incomplete"] == (state not in {"complete", "stale"})
+    if state in {"complete", "stale"}:
+        assert row["nb_files"] == 1
+        assert row["size_bytes"] == len(b"completed model data")
+    assert {path.name: path.read_bytes() for path in blobs.iterdir()} == before

tests/test_database_test_isolation.py

@@ -85,6 +85,9 @@ def test_actual_tests_restore_process_state_and_ignore_inherited_database(tmp_pa
         import os
         import sys
         import pytest
+        import tempfile
+        from pathlib import Path
+        from sqlalchemy.engine import make_url
 
         def snapshot():
             import core
@@ -107,7 +110,11 @@ def test_actual_tests_restore_process_state_and_ignore_inherited_database(tmp_pa
         class StateGuard:
             def pytest_sessionstart(self):
                 self.initial = snapshot()
-                assert self.initial[0] == "sqlite:///:memory:"
+                assert self.initial[0] != inherited_url
+                self.default_path = Path(make_url(self.initial[0]).database)
+                assert self.default_path.is_file()
+                assert self.default_path.name == "collection.db"
+                assert self.default_path.parent == Path(tempfile.gettempdir()).parent
 
             def pytest_collection_finish(self):
                 assert snapshot() == self.initial, "collection changed database bindings"
@@ -118,9 +125,11 @@ def test_actual_tests_restore_process_state_and_ignore_inherited_database(tmp_pa
                 assert snapshot() == self.initial, "test leaked database or module state"
 
         inherited_url = os.environ["DATABASE_URL"]
+        guard = StateGuard()
         result = pytest.main(["-q", "-p", "no:cacheprovider", *sys.argv[1:]],
-                             plugins=[StateGuard()])
+                             plugins=[guard])
         assert os.environ["DATABASE_URL"] == inherited_url
+        assert not guard.default_path.exists(), "collection database outlived its runtime"
         raise SystemExit(result)
     ''')
     result = subprocess.run(

tests/test_delete_user_invalidates_token_cache.py

@@ -11,6 +11,8 @@ cache from the DB, where the rows are already gone, and the token is rejected.
 """
 import asyncio
 import types
+import pytest
+from fastapi import HTTPException
 
 from routes.auth_routes import setup_auth_routes, DeleteUserRequest
 
@@ -60,12 +62,9 @@ def test_refused_delete_does_not_invalidate_cache():
     invalidations = []
     router = setup_auth_routes(_auth_manager(delete_result=False))
     handler = _handler(router)
-    try:
+    with pytest.raises(HTTPException) as refused:
         asyncio.run(handler(DeleteUserRequest(username="admin"), _fake_request(invalidations)))
-        raised = False
+    assert refused.value.status_code == 400
-    except Exception:
-        raised = True
-    assert raised, "a refused delete should raise (HTTP 400)"
     assert invalidations == [], "a refused delete must not touch the token cache"
 
 

tests/test_document_export_zip_ids.py

@@ -0,0 +1,47 @@
+"""POST /api/documents/export-zip answers 400 for ids that are not a list of document ids."""
+
+import json
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import pytest
+from fastapi import HTTPException
+
+from tests.helpers.import_state import clear_fake_database_modules
+
+clear_fake_database_modules()
+
+import routes.document_routes as droutes
+
+
+class _Req:
+    def __init__(self, body):
+        self.state = SimpleNamespace(current_user="tester")
+        self._body = body
+
+    async def json(self):
+        return self._body
+
+
+def _export_zip():
+    router = droutes.setup_document_routes(MagicMock(), None)
+    for r in router.routes:
+        if getattr(r, "path", None) == "/api/documents/export-zip":
+            return r.endpoint
+    raise RuntimeError("export-zip route not found")
+
+
+@pytest.mark.parametrize("body", [[], "x", 5, {}, {"ids": []}, {"ids": None}])
+async def test_missing_ids_is_a_400(body):
+    with pytest.raises(HTTPException) as exc:
+        await _export_zip()(_Req(body))
+    assert exc.value.status_code == 400
+    assert exc.value.detail == "No documents specified"
+
+
+@pytest.mark.parametrize("ids", [5, "abc", {"a": 1}, [1, 2], [["a"]], ["a", None]])
+async def test_ids_that_are_not_a_list_of_strings_is_a_400(ids):
+    with pytest.raises(HTTPException) as exc:
+        await _export_zip()(_Req({"ids": ids}))
+    assert exc.value.status_code == 400
+    assert exc.value.detail == "ids must be a list of document ids"

tests/test_editor_photoshop_features_more4.py

@@ -93,7 +93,7 @@ def test_custom_filter_reuses_convolve_kernel():
     assert "convolve(readSource(snap), k, divisor, bias, 1)" in EFFECTS
 
 
-def test_new_filter_aliases_registered():
+def test_blur_and_custom_filter_aliases_registered():
     assert "'box-blur': 'blur-box'" in EDITOR
     assert "'surface-blur': 'blur-surface'" in EDITOR
     assert "custom: 'other-custom'" in EDITOR

tests/test_email_linkify_security_js.py

@@ -1,6 +1,7 @@
 """DOM-XSS regressions for email plain-text linkification helpers."""
 
 import json
+import os
 import shutil
 import subprocess
 import textwrap
@@ -100,3 +101,55 @@ def test_email_html_sanitizer_runs_to_fixpoint():
     assert "for (let i = 0; i < 4; i++)" in source
     assert "const next = _sanitizeHtmlOnce(out);" in source
     assert "if (next === out) break;" in source
+
+
+@pytest.mark.skipif(not _HAS_NODE, reason="node binary not on PATH")
+def test_linkify_preserves_text_and_link_targets_in_a_real_dom():
+    cases = [
+        ("See https://example.test/contact?email=jane@example.test&next=1 and bob@example.test.",
+         ["https://example.test/contact?email=jane@example.test&next=1", "mailto:bob@example.test"]),
+        ("https://alice@example.test/path", ["https://alice@example.test/path"]),
+        ("www.example.test/?email=jane@example.test", ["https://www.example.test/?email=jane@example.test"]),
+        ("<https://example.test/path> and \"jane@example.test\"",
+         ["https://example.test/path", "mailto:jane@example.test"]),
+        ("(https://example.test/path), jane@example.test!",
+         ["https://example.test/path", "mailto:jane@example.test"]),
+        ("https://example.test/?literal=&quot;&next=1", ["https://example.test/?literal=&quot;&next=1"]),
+        ('<img src=x onerror="alert(1)"> jane@example.test', ["mailto:jane@example.test"]),
+        ("No links\njust <plain> & text", []),
+    ]
+    script = r'''
+      import { chromium } from 'playwright';
+      const browser = await chromium.launch({ headless: true });
+      try {
+        const page = await browser.newPage();
+        const origin = process.env.ODYSSEUS_TEST_STATIC_ORIGIN;
+        await page.goto(`${origin}/static/js/documentStats.js`);
+        const results = await page.evaluate(async (cases) => {
+          const { _escLinkify } = await import('/static/js/emailLibrary/utils.js');
+          return cases.map(([raw]) => {
+            const div = document.createElement('div');
+            div.innerHTML = _escLinkify(raw);
+            return {
+              text: div.textContent,
+              hrefs: [...div.querySelectorAll('a')].map(a => a.getAttribute('href')),
+              otherElements: [...div.querySelectorAll('*')].map(e => e.tagName).filter(t => t !== 'A'),
+              unsafeBlankLink: [...div.querySelectorAll('a[target="_blank"]')]
+                .some(a => !a.relList.contains('noopener') || !a.relList.contains('noreferrer')),
+            };
+          });
+        }, JSON.parse(process.env.ODYSSEUS_LINKIFY_CASES));
+        console.log(JSON.stringify(results));
+      } finally {
+        await browser.close();
+      }
+    '''
+    result = subprocess.run(
+        ["node", "--input-type=module", "-e", script], cwd=_REPO,
+        capture_output=True, text=True, timeout=30,
+        env={**os.environ, "ODYSSEUS_LINKIFY_CASES": json.dumps(cases)},
+    )
+    assert result.returncode == 0, result.stderr
+    results = json.loads(result.stdout)
+    for (raw, hrefs), actual in zip(cases, results, strict=True):
+        assert actual == {"text": raw, "hrefs": hrefs, "otherElements": [], "unsafeBlankLink": False}, raw

tests/test_email_list_stale_while_revalidate.py

@@ -56,14 +56,35 @@ def test_cold_start_with_no_snapshot_and_no_index_is_not_marked_stale(monkeypatc
     # falls through to the real (blocking) path, so the response is not a
     # stale-marked snapshot.
     endpoint = _list_endpoint()
+    calls = []
+
+    class EmptyMailbox:
+        def select(self, folder, readonly=False):
+            calls.append(("select", folder, readonly))
+            return "OK", [b"0"]
+
+        def uid(self, command, *args):
+            assert command == "SEARCH"
+            calls.append(("search", args))
+            return "OK", [b""]
+
+        def logout(self):
+            pass
+
+    def connect(*args, **kwargs):
+        calls.append(("connect", args, kwargs))
+        return EmptyMailbox()
+
+    monkeypatch.setattr(er, "_start_poller", lambda: None)
     monkeypatch.setattr(er, "_email_index_list", lambda *a, **k: ([], 0, None))
-    monkeypatch.setattr(er, "_imap_connect", lambda *a, **k: (_ for _ in ()).throw(
+    monkeypatch.setattr(er, "_imap_connect", connect)
-        RuntimeError("offline test")))
+    result = _call(endpoint)
-    try:
+    assert [call[0] for call in calls] == ["connect", "select", "search"]
-        result = _call(endpoint)
+    assert calls[1][2] is True
-    except Exception:
-        return  # the cold path legitimately attempted real IMAP
     assert isinstance(result, dict)
+    assert result["emails"] == []
+    assert result["total"] == 0
+    assert not result.get("error")
     assert not (result.get("sync") or {}).get("stale")
 
 

tests/test_email_paperclip_files_browser.py

@@ -0,0 +1,61 @@
+"""The native FileList must survive resetting the compose file input."""
+import json
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+
+from tests.helpers.document_source import function_body
+
+
+@pytest.mark.parametrize("handler,uploader", [("_handleAttachUpload", "_uploadComposeFiles"), ("_handleMarkdownImageUpload", "_uploadMarkdownImages")])
+@pytest.mark.parametrize("selections", [[["one.png"]], [["one.png", "two.png"]], [["one.png"], ["one.png"]], [[]]])
+def test_compose_file_selection_survives_reset(tmp_path, selections, handler, uploader):
+    if not shutil.which("node"):
+        pytest.skip("node not installed")
+    repo = Path(__file__).resolve().parents[1]
+    for name, icon in [("one.png", "icon-192.png"), ("two.png", "icon-512.png")]:
+        shutil.copyfile(repo / "static/icons" / icon, tmp_path / name)
+    script = r"""
+const fs=require('node:fs'),assert=require('node:assert/strict'),crypto=require('node:crypto');
+let chromium;try{({chromium}=require('playwright'));}catch(e){if(e.code==='MODULE_NOT_FOUND')process.exit(77);throw e;}
+(async()=>{
+ const browser=await chromium.launch({headless:true});
+ try{
+  const page=await browser.newPage();const errors=[];page.on('pageerror',e=>errors.push(e.message));
+  await page.goto(process.env.ODYSSEUS_TEST_STATIC_ORIGIN+'/static/js/storage.js');
+  await page.setContent('<input type="file" multiple aria-label="Compose attachments">');
+  await page.addScriptTag({content:`
+    window.uploads=[];window.finished=0;
+    async function __UPLOADER__(files){
+      for(const file of files){
+        const hash=Array.from(new Uint8Array(await crypto.subtle.digest('SHA-256',await file.arrayBuffer()))).map(x=>x.toString(16).padStart(2,'0')).join('');
+        window.uploads.push({name:file.name,hash});
+      }
+    }
+    __HANDLER__
+    document.querySelector('input').addEventListener('change',async event=>{
+      await __HANDLER_NAME__(event);window.finished++;
+    });
+  `});
+  const selections=JSON.parse(process.argv[1]),root=process.argv[2],expected=[];
+  for(const [index,names] of selections.entries()){
+   await page.getByLabel('Compose attachments').setInputFiles(names.map(name=>root+'/'+name));
+   await page.waitForFunction(n=>window.finished===n,index+1);
+   expected.push(...names.map(name=>({name,hash:crypto.createHash('sha256').update(fs.readFileSync(root+'/'+name)).digest('hex')})));
+   assert.deepEqual(await page.evaluate(()=>window.uploads),expected);
+   assert.equal(await page.getByLabel('Compose attachments').inputValue(),'');
+   assert.equal(await page.getByLabel('Compose attachments').evaluate(el=>el.files.length),0);
+  }
+  assert.deepEqual(errors,[]);console.log(JSON.stringify({files:expected.length}));
+ }finally{await browser.close();}
+})().catch(e=>{console.error(e);process.exitCode=1;});
+""".replace("__HANDLER__", function_body(handler)).replace("__HANDLER_NAME__", handler).replace("__UPLOADER__", uploader)
+    result = subprocess.run(["node", "-e", script, json.dumps(selections), str(tmp_path)],
+                            cwd=Path(__file__).resolve().parents[1], capture_output=True,
+                            text=True, timeout=30)
+    if result.returncode == 77:
+        pytest.skip("playwright not installed")
+    assert result.returncode == 0, result.stdout + result.stderr
+    assert json.loads(result.stdout)["files"] == sum(map(len, selections))

tests/test_font_routes.py

@@ -1,3 +1,8 @@
+import pytest
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+import routes.font_routes as font_routes
 from routes.font_routes import _derive_family
 
 
@@ -9,3 +14,37 @@ def test_derive_family_splits_common_family_suffixes():
     assert _derive_family("FiraCode-SemiBold.ttf") == "Fira Code"
     assert _derive_family("NotoSans-Bold.otf") == "Noto Sans"
     assert _derive_family("RobotoSlab-Bold.woff2") == "Roboto Slab"
+
+
+@pytest.mark.parametrize("filename,expected", [
+    ("JetBrainsMono-SemiBoldItalic.woff2", "JetBrains Mono"),
+    ("FiraCode-BoldItalic.ttf", "Fira Code"),
+    ("NotoSans-ExtraLightItalic.otf", "Noto Sans"),
+    ("RobotoSlab-UltraBoldOblique.WOFF2", "Roboto Slab"),
+    ("SourceSansPro-Bold-Italic.woff", "Source Sans Pro"),
+    ("FiraCode-SemiBold_Italic.ttf", "Fira Code"),
+    ("FiraCode-Italic.woff2", "Fira Code"),
+    ("FiraCode-VF.woff2", "Fira Code"),
+    ("MyFont-Book.woff2", "My Font Book"),
+])
+def test_compound_weight_and_style_are_not_part_of_the_family(filename, expected):
+    assert _derive_family(filename) == expected
+
+
+def test_custom_font_endpoint_groups_all_styles_under_one_family(tmp_path, monkeypatch):
+    files = ["JetBrainsMono-Regular.woff2", "JetBrainsMono-SemiBoldItalic.woff2",
+             "JetBrainsMono-Bold-Oblique.ttf"]
+    for name in [*files, "OtherFont-Regular.otf", "README.md"]:
+        (tmp_path / name).write_bytes(b"filename-only discovery fixture")
+    monkeypatch.setattr(font_routes, "CUSTOM_FONTS_DIR", str(tmp_path))
+    app = FastAPI()
+    app.include_router(font_routes.setup_font_routes())
+    with TestClient(app) as client:
+        response = client.get("/api/fonts/custom")
+    assert response.status_code == 200
+    fonts = response.json()["fonts"]
+    assert set(fonts) == {"JetBrains Mono", "Other Font"}
+    assert fonts["JetBrains Mono"] == [
+        {"file": name, "url": f"/static/fonts/custom/{name}", "format": name.rsplit(".", 1)[1]}
+        for name in sorted(files)
+    ]

tests/test_form_markdown_roundtrip.py

@@ -8,6 +8,8 @@ label with `[^*]+`, so they could not match a label like "Email *" / "State *"
 stayed empty and the exported PDF (and signed-reply attachment) came out blank
 for that field, with no error.
 """
+import pytest
+
 from src.pdf_form_doc import render_form_as_markdown, parse_markdown_to_values
 
 
@@ -38,3 +40,38 @@ def test_plain_labels_and_colon_values_unaffected():
     vals = parse_markdown_to_values(md)
     assert vals["name"] == "Alice"
     assert vals["time"] == "9:00 sharp"
+
+
+@pytest.mark.parametrize("name", ["Prénom", "Año", "İsim", "氏名", "اسم", "n²", "Cafe\u0301"])
+@pytest.mark.parametrize("field_type, value", [
+    ("text", "Alice"), ("checkbox", True), ("choice", "Yes"),
+])
+def test_unicode_field_names_survive_export_roundtrip(name, field_type, value):
+    fields = [{"name": name, "label": "Field *", "type": field_type,
+               "value": value, "options": ["Yes", "No"], "page": 1}]
+    markdown = render_form_as_markdown(fields, "upload", "Form")
+    marker_name = markdown.split("<!-- field=", 1)[1].split(" type=", 1)[0]
+    assert marker_name.isascii()
+    assert parse_markdown_to_values(markdown) == {name: value}
+
+
+def test_previously_saved_unicode_markers_remain_exportable():
+    # Older imports emitted Unicode letters literally. Their persisted
+    # markdown must also work after upgrading, without re-importing the PDF.
+    markdown = (
+        "- **First name:** Alice <!-- field=Prénom type=text -->\n"
+        "- [x] **Consent** <!-- field=同意 type=checkbox -->\n"
+        "- **Year** [2025 / 2026]: 2026 <!-- field=Año type=choice -->\n"
+    )
+    assert parse_markdown_to_values(markdown) == {
+        "Prénom": "Alice", "同意": True, "Año": "2026",
+    }
+
+
+def test_unicode_names_do_not_collide_with_literal_percent_or_ascii_names():
+    names = ["Prénom", "Pr%C3%A9nom", "Prenom", "Section.1_name-2"]
+    fields = [{"name": name, "type": "text", "value": str(i), "page": 1}
+              for i, name in enumerate(names)]
+    markdown = render_form_as_markdown(fields, "upload", "Form")
+    assert "field=Section.1_name-2 " in markdown
+    assert parse_markdown_to_values(markdown) == {name: str(i) for i, name in enumerate(names)}

tests/test_gallery_image_bad_input.py

@@ -0,0 +1,46 @@
+import asyncio
+import base64
+import io
+
+import pytest
+from fastapi import HTTPException
+from PIL import Image
+
+import routes.gallery_routes as gallery
+
+ROUTES = ['sharpen', 'remove-bg', 'denoise', 'upscale-local', 'enhance-face']
+
+
+def route(name):
+    return next(r for r in gallery.setup_gallery_routes().routes if r.path == f'/api/image/{name}')
+
+
+def call(name, monkeypatch, body):
+    monkeypatch.setattr(gallery, 'require_privilege', lambda *args: 'alice')
+    monkeypatch.setattr(gallery, '_local_image_models_enabled', lambda: True)
+    monkeypatch.setattr(gallery, '_remote_image_ops_base', lambda: '')
+
+    class Request:
+        async def json(self):
+            return body
+
+    return asyncio.run(route(name).endpoint(Request()))
+
+
+@pytest.mark.parametrize('name', ROUTES)
+@pytest.mark.parametrize('body', [
+    {},
+    {'image': '!!notb64'},
+    {'image': base64.b64encode(b'hello').decode()},
+])
+def test_bad_image_is_a_client_error(name, body, monkeypatch):
+    with pytest.raises(HTTPException) as exc:
+        call(name, monkeypatch, body)
+    assert exc.value.status_code == 400
+
+
+def test_sharpen_still_accepts_a_valid_image(monkeypatch):
+    output = io.BytesIO()
+    Image.new('RGB', (4, 4), 'red').save(output, format='PNG')
+    result = call('sharpen', monkeypatch, {'image': base64.b64encode(output.getvalue()).decode()})
+    assert Image.open(io.BytesIO(base64.b64decode(result['image']))).size == (4, 4)

tests/test_generate_image_native_schema.py

@@ -0,0 +1,59 @@
+"""Native image calls must match the server schema and provide a prompt.
+
+A missing prompt must be rejected before the fallback line parser mistakes the
+whole arguments object for the requested image.
+"""
+
+import json
+
+
+def _schemas():
+    import src.agent_tools  # noqa: F401  (tool_schemas <-> agent_tools import cycle)
+    from src.tool_schemas import FUNCTION_TOOL_SCHEMAS
+
+    return {s["function"]["name"]: s["function"] for s in FUNCTION_TOOL_SCHEMAS}
+
+
+def _convert(arguments):
+    import src.agent_tools  # noqa: F401
+    from src.tool_schemas import function_call_to_tool_block
+
+    return function_call_to_tool_block("generate_image", json.dumps(arguments))
+
+
+def test_generate_image_is_offered_to_native_function_calling_models():
+    schemas = _schemas()
+    assert "generate_image" in schemas, (
+        "generate_image has no native schema, so it is filtered out of the tools "
+        "sent to API models even when the selector picks it"
+    )
+    assert schemas["generate_image"]["parameters"]["required"] == ["prompt"]
+
+
+def test_native_call_reaches_the_executor_with_its_arguments_intact():
+    from src.tool_execution import _build_mcp_args
+
+    args = {"prompt": "a cat riding a bicycle", "model": "gpt-image-1",
+            "size": "1024x1024", "quality": "high"}
+    block = _convert(args)
+    assert block is not None and block.tool_type == "generate_image"
+    assert _build_mcp_args("generate_image", block.content) == args
+
+
+def test_call_without_a_prompt_is_rejected_instead_of_drawing_its_own_arguments():
+    """No prompt key means the line parser takes the whole JSON blob as the prompt,
+    so the model gets an image of its own arguments."""
+    assert _convert({"size": "512x512"}) is None
+    assert _convert({}) is None
+    assert _convert({"prompt": "   "}) is None
+
+
+def test_advertised_parameters_are_the_ones_the_image_server_accepts():
+    """Anything the schema advertises but the executor drops is a silent no-op."""
+    from mcp_servers.image_gen_server import list_tools
+    import asyncio
+
+    server_schema = asyncio.run(list_tools())[0].inputSchema
+    assert set(_schemas()["generate_image"]["parameters"]["properties"]) == set(
+        server_schema["properties"]
+    )

tests/test_group_peer_history_identity.py

@@ -0,0 +1,67 @@
+"""Peer injections retain their context role but hydrate as the named agent."""
+import json
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+from tests.helpers.js_modules import js_function_source
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def _node(script, *args):
+    result = subprocess.run(['node', '--input-type=module', '-e', script, *map(json.dumps, args)],
+                            cwd=ROOT, capture_output=True, text=True, timeout=15)
+    assert result.returncode == 0, result.stderr
+    return json.loads(result.stdout)
+
+
+@pytest.mark.skipif(not shutil.which('node'), reason='node not installed')
+@pytest.mark.parametrize('mode', ['parallel', 'round-robin'])
+def test_peer_sync_persists_explicit_identity(mode):
+    source = (ROOT / 'static/js/group.js').read_text()
+    functions = '\n'.join(js_function_source(name, source) for name in ['_syncAllResponses', '_sendRoundRobin'])
+    data = _node(r'''
+const source=JSON.parse(process.argv[1]),mode=JSON.parse(process.argv[2]);
+const _models=[{mid:'model-a',display:'Model A',_groupName:'Persona A'}, {mid:'model-b',display:'Model B'}];
+const _participantSessions=['a','b'],API_BASE='http://app',sent=[];
+let _abortControllers=[];
+const uiModule={scrollHistory(){}};
+function _createGroupBubble(model){return {dataset:{raw:'Reply from '+model.mid}};}
+async function _streamToHolder(){} function _saveState(){}
+async function fetch(url,options){sent.push({url,...JSON.parse(options.body)});return {ok:true};}
+eval(source+'\nglobalThis.runSync=async()=> mode==="parallel" ? _syncAllResponses(_models.map(_createGroupBubble)) : _sendRoundRobin("Hi",null);');
+await globalThis.runSync();console.log(JSON.stringify(sent));
+''', functions, mode)
+    assert len(data) == 2
+    for item in data:
+        message = item['messages'][0]
+        assert message['role'] == 'user'  # Model context must retain peer attribution.
+        peer = 'model-b' if item['url'].endswith('/a/inject_messages') else 'model-a'
+        name = 'Model B' if peer == 'model-b' else 'Persona A'
+        assert message['content'] == f'[{name}]: Reply from {peer}'
+        assert message['metadata'] == {'group_peer': True, 'group_model': name, 'model': peer}
+
+
+@pytest.mark.skipif(not shutil.which('node'), reason='node not installed')
+@pytest.mark.parametrize('role,content,meta,expected_role,expected_content', [
+    ('user', '[Persona A]: Reply', {'group_peer': True, 'group_model': 'Persona A'}, 'assistant', 'Reply'),
+    ('user', '[Persona A]: Human text', {}, 'user', '[Persona A]: Human text'),
+    ('user', 'Human text', {}, 'user', 'Human text'),
+    ('assistant', 'Parent reply', {'group_model': 'Persona A'}, 'assistant', 'Parent reply'),
+    ('user', 'No matching prefix', {'group_peer': True, 'group_model': 'Persona A'}, 'user', 'No matching prefix'),
+    ('user', '[3]: Human text', {'group_peer': True, 'group_model': 3}, 'user', '[3]: Human text'),
+    ('user', '[Persona A]: Human text', {'group_peer': False, 'group_model': 'Persona A'}, 'user', '[Persona A]: Human text'),
+])
+def test_history_uses_explicit_peer_identity_only(role, content, meta, expected_role, expected_content):
+    source = js_function_source('_renderHistoryMessage', (ROOT / 'static/js/sessions.js').read_text())
+    result = _node(r'''
+const source=JSON.parse(process.argv[1]),message=JSON.parse(process.argv[2]);
+function _displayHistoryContent(text){return text;}function _stripUserVisionBlocks(text){return text;}
+function _addHistoryMessageWithFullRenderer(role,content,model,metadata){return {role,content,metadata};}
+const run=eval('('+source+')');console.log(JSON.stringify(run(message,'session-model')));
+''', source, {'role': role, 'content': content, 'metadata': meta})
+    assert result['role'] == expected_role
+    assert result['content'] == expected_content
+    assert result['metadata'] == {**meta, '_fromHistory': True}

tests/test_hwfit_hf_discovery_moe.py

@@ -0,0 +1,33 @@
+from services.hwfit.hf_discovery import _entry_from_collection_item
+
+_SOURCE = {"owner": "Qwen", "provider": "Qwen"}
+_COLLECTION = {"title": "Qwen3.6"}
+
+
+def _item(repo_id, num_params):
+    item = {"id": repo_id, "type": "model", "pipeline_tag": "text-generation"}
+    if num_params is not None:
+        item["numParameters"] = num_params
+    return item
+
+
+def test_moe_active_size_is_read_from_name_when_hf_reports_total():
+    # HF collection items carry numParameters for nearly every repo. The MoE
+    # active size only exists in the name, so it must still be parsed.
+    entry = _entry_from_collection_item(_COLLECTION, _item("Qwen/Qwen3.6-35B-A3B", 35_951_822_704), _SOURCE)
+    assert entry["parameters_raw"] == 35_951_822_704
+    assert entry["is_moe"] is True
+    assert entry["active_parameters"] == 3_000_000_000
+
+
+def test_moe_entry_without_hf_count_falls_back_to_name():
+    entry = _entry_from_collection_item(_COLLECTION, _item("Qwen/Qwen3-Coder-30B-A3B-Instruct", None), _SOURCE)
+    assert entry["parameters_raw"] == 30_000_000_000
+    assert entry["is_moe"] is True
+    assert entry["active_parameters"] == 3_000_000_000
+
+
+def test_dense_entry_keeps_hf_count_and_is_not_moe():
+    entry = _entry_from_collection_item(_COLLECTION, _item("Qwen/Qwen3.6-27B", 27_781_427_952), _SOURCE)
+    assert entry["parameters_raw"] == 27_781_427_952
+    assert "is_moe" not in entry

tests/test_hwfit_hf_discovery_pipeline_filter.py

@@ -0,0 +1,31 @@
+import pytest
+
+from services.hwfit.hf_discovery import _entry_from_collection_item
+
+_SOURCE = {"owner": "google", "provider": "Google"}
+_COLLECTION = {"title": "collection"}
+
+
+@pytest.mark.parametrize("repo_id,pipeline_tag", [
+    ("google/timesfm-3.0-pytorch", "time-series-forecasting"),
+    ("google/siglip2-base-patch16-224", "zero-shot-image-classification"),
+    ("moonshotai/MoonViT-SO-400M", "image-feature-extraction"),
+    ("zai-org/CogView4-6B", "text-to-image"),
+    ("zai-org/CogVideoX-5b-I2V", "image-to-video"),
+    ("zai-org/CogVideoX1.5-5B", "text-to-video"),
+])
+def test_non_llm_collection_models_are_skipped(repo_id, pipeline_tag):
+    item = {"id": repo_id, "type": "model", "pipeline_tag": pipeline_tag, "numParameters": 1_000_000_000}
+    assert _entry_from_collection_item(_COLLECTION, item, _SOURCE) is None
+
+
+@pytest.mark.parametrize("repo_id,pipeline_tag", [
+    ("google/gemma-4-12B-it", "image-text-to-text"),
+    ("Qwen/Qwen3.6-27B", "text-generation"),
+    ("Qwen/Qwen3-Embedding-4B", "feature-extraction"),
+    ("NousResearch/DeepHermes-Egregore-v2-RLAIF-8b-Atropos", "reinforcement-learning"),
+    ("Qwen/Qwen3-Next-80B-A3B-Instruct", ""),
+])
+def test_servable_collection_models_are_kept(repo_id, pipeline_tag):
+    item = {"id": repo_id, "type": "model", "pipeline_tag": pipeline_tag, "numParameters": 4_000_000_000}
+    assert _entry_from_collection_item(_COLLECTION, item, _SOURCE) is not None

tests/test_image_quality_defaults.py

@@ -0,0 +1,74 @@
+"""Explicit quality choices must take precedence over the configured default."""
+import asyncio
+import json
+
+import pytest
+from PIL import Image
+
+
+@pytest.mark.parametrize("caller", ["json", "lines", "mcp", "edit"])
+@pytest.mark.parametrize("requested,configured,expected", [
+    ("medium", "high", "medium"),
+    ("medium", "low", "medium"),
+    ("low", "high", "low"),
+    ("high", "low", "high"),
+    ("auto", "high", "auto"),
+    (None, "high", "high"),
+    (None, None, "medium"),
+])
+def test_image_quality_precedence(monkeypatch, tmp_path, caller, requested, configured, expected):
+    from src import ai_interaction, settings
+    from mcp_servers.image_gen_server import call_tool
+
+    config = {"image_gen_enabled": True, "image_model": "gpt-image-1"}
+    if configured is not None:
+        config["image_quality"] = configured
+    monkeypatch.setattr(settings, "load_settings", lambda: config)
+    monkeypatch.setattr(ai_interaction, "_resolve_model", lambda *args, **kwargs:
+                        ("http://fixture.invalid/v1/chat/completions", "gpt-image-1", {}))
+    captured = []
+
+    class Response:
+        status_code = 418
+        text = "Fixture captured request"
+
+        def json(self):
+            return {"error": {"message": self.text}}
+
+    class Client:
+        def __init__(self, *args, **kwargs):
+            pass
+
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, *args):
+            pass
+
+        async def post(self, url, **kwargs):
+            captured.append(kwargs.get("json") or kwargs.get("data"))
+            return Response()
+
+    monkeypatch.setattr("httpx.AsyncClient", Client)
+    args = {"prompt": "Fixture image", "model": "gpt-image-1"}
+    if requested is not None:
+        args["quality"] = requested
+    if caller == "json":
+        operation = ai_interaction.do_generate_image(json.dumps(args), incognito=True)
+    elif caller == "lines":
+        content = "Fixture image\ngpt-image-1\n1024x1024"
+        if requested is not None:
+            content += "\n" + requested
+        operation = ai_interaction.do_generate_image(content, incognito=True)
+    elif caller == "mcp":
+        operation = call_tool("generate_image", args)
+    else:
+        source = tmp_path / "fixture.png"
+        Image.new("RGB", (2, 2), "white").save(source)
+        kwargs = {} if requested is None else {"quality": requested}
+        operation = ai_interaction.do_edit_image("Fixture image", str(source), size="1024x1024",
+                                               incognito=True, **kwargs)
+    result = asyncio.run(operation)
+    assert "418" in str(result), result
+    assert len(captured) == 1
+    assert captured[0]["quality"] == expected

tests/test_integrations_store_shape.py

@@ -177,3 +177,28 @@ def test_update_integration_rejects_invalid_base_url_without_changing_existing(
     assert exc.value.status_code == 400
     assert exc.value.detail == message
     assert integrations.load_integrations()[0]["base_url"] == "https://example.test"
+
+
+@pytest.mark.parametrize("body", [[], "x", 5, None])
+def test_create_integration_rejects_non_object_body(integrations_routes, body):
+    endpoint, session_cookie, http_exception = integrations_routes
+    create_integration = endpoint("/api/auth/integrations", "POST")
+
+    with pytest.raises(http_exception) as exc:
+        asyncio.run(create_integration(_JsonRequest(body, session_cookie)))
+
+    assert exc.value.status_code == 400
+    assert integrations.load_integrations() == []
+
+
+@pytest.mark.parametrize("body", [[], "x", 5, None])
+def test_update_integration_rejects_non_object_body(integrations_routes, body):
+    endpoint, session_cookie, http_exception = integrations_routes
+    update_integration = endpoint("/api/auth/integrations/{integration_id}", "PUT")
+    integrations.save_integrations([{"id": "existing", "name": "Original", "base_url": "https://example.test"}])
+
+    with pytest.raises(http_exception) as exc:
+        asyncio.run(update_integration("existing", _JsonRequest(body, session_cookie)))
+
+    assert exc.value.status_code == 400
+    assert integrations.load_integrations()[0]["name"] == "Original"

tests/test_llm_core_fallback.py

@@ -1091,6 +1091,43 @@ def test_degenerate_stream_guard_allows_long_rows_with_changing_values():
     assert guard.check(rows) is None
 
 
+def _latex_matrix(cell):
+    rows = (" & ".join(cell(i, j) for j in range(8)) for i in range(8))
+    return "$$\\begin{pmatrix}\n" + " \\\\\n".join(rows) + "\n\\end{pmatrix}$$"
+
+
+@pytest.mark.parametrize("text", [
+    _latex_matrix(lambda i, j: "0"),
+    _latex_matrix(lambda i, j: "1" if i == j else "0"),
+    "buf = [" + ", ".join("0" for _ in range(64)) + "]",
+])
+@pytest.mark.parametrize("chunk", [1, 6, 64])
+def test_degenerate_stream_guard_allows_numeric_data(text, chunk):
+    """#5560: zero/identity matrices and padded buffers repeat numbers by
+    design; they are not token collapse."""
+    guard = llm_core._DegenerateStreamGuard("math-model")
+
+    assert all(guard.check(text[i:i + chunk]) is None for i in range(0, len(text), chunk))
+
+
+def test_degenerate_stream_guard_still_stops_an_endless_number_run():
+    guard = llm_core._DegenerateStreamGuard("looping-model")
+
+    chunk = guard.check("0 " * 3000)
+
+    assert chunk is not None
+    assert "numbers in a row" in chunk
+
+
+def test_degenerate_stream_guard_still_stops_a_mixed_word_number_loop():
+    guard = llm_core._DegenerateStreamGuard("looping-model")
+
+    chunk = guard.check("row 0 0 0 0 " * 300)
+
+    assert chunk is not None
+    assert "repeated" in chunk
+
+
 def test_terminal_stream_retries_degenerate_generation_before_emitting(monkeypatch):
     calls = 0
     requests = []

tests/test_manage_tasks_cron.py

@@ -51,7 +51,7 @@ async def test_cron_create_edit_resume_and_invalid_edit_rollback(storage):
 
 
 @pytest.mark.asyncio
-@pytest.mark.parametrize('expression', [None, '', 'bad', '99 99 * * *'])
+@pytest.mark.parametrize('expression', [None, '', 'bad', '99 99 * * *', '* * * * * *', '* * * * * * *'])
 async def test_invalid_cron_does_not_create_inert_task(storage, expression):
     result = await call(action='create', name='Invalid', prompt='Test', schedule='cron',
                         cron_expression=expression)
@@ -171,3 +171,21 @@ async def test_invalid_cron_retime_rolls_back_all_edits(storage):
         row = db.get(database.ScheduledTask, result['task_id'])
         assert row.name == 'Digest' and row.cron_expression == '15 9 * * 1,3,5'
         assert row.next_run == datetime(2026, 10, 5, 9, 15)
+
+
+@pytest.mark.parametrize('expression', ['* * * * *', '15 9 * * 1,3,5', '*/5 * * * *'])
+def test_is_valid_cron_accepts_five_fields(expression):
+    from src.task_scheduler import is_valid_cron
+    assert is_valid_cron(expression)
+
+
+@pytest.mark.parametrize('expression', [None, '', 'bad', '99 99 * * *', '* * * * * *', '* * * * * * *', '* * * *'])
+def test_is_valid_cron_rejects_everything_else(expression):
+    from src.task_scheduler import is_valid_cron
+    assert not is_valid_cron(expression)
+
+
+@pytest.mark.parametrize('expression', ['* * * * * *', '* * * * * * *'])
+def test_compute_next_run_refuses_a_seconds_field(expression):
+    from src.task_scheduler import compute_next_run
+    assert compute_next_run('cron', '09:00', None, None, cron_expression=expression) is None

tests/test_mcp_manager.py

@@ -39,3 +39,74 @@ def test_http_transport_routes_to_start_http_connect():
         result = asyncio.run(mgr.connect_server("id1", "n", "http", url="https://x/mcp"))
     assert result == "ROUTED"
     m.assert_called_once()
+
+
+def _patch_http_transport(monkeypatch, initialize):
+    """Fake the Streamable HTTP transport and record, for each close, whether
+    it ran in the task that opened it (AnyIO requires that)."""
+    import contextlib
+    import mcp
+    import mcp.client.streamable_http as streamable_http
+    import src.mcp_oauth as mcp_oauth
+
+    closes = []
+
+    @contextlib.asynccontextmanager
+    async def fake_client(url, auth=None):
+        opener = asyncio.current_task()
+        try:
+            yield (object(), object(), lambda: None)
+        finally:
+            closes.append(asyncio.current_task() is opener)
+
+    class FakeSession:
+        def __init__(self, *a):
+            pass
+
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, *exc):
+            return False
+
+    FakeSession.initialize = initialize
+    monkeypatch.setattr(streamable_http, "streamablehttp_client", fake_client)
+    monkeypatch.setattr(mcp, "ClientSession", FakeSession)
+    monkeypatch.setattr(mcp_oauth, "build_provider", lambda *a, **k: None)
+    return closes
+
+
+def test_failed_http_connect_closes_transport_in_its_own_task(monkeypatch):
+    """#5518: a failed connect left its transport for the GC, which exited the
+    AnyIO cancel scope from the wrong task and kept the event loop spinning."""
+    async def initialize(self):
+        raise RuntimeError("Session terminated")
+
+    closes = _patch_http_transport(monkeypatch, initialize)
+    mgr = McpManager()
+
+    ok = asyncio.run(mgr._start_http_connect("s1", "remote", "http://x/mcp/", wait=2))
+
+    assert ok is False
+    assert mgr._connections["s1"]["status"] == "error"
+    assert "Session terminated" in mgr._connections["s1"]["error"]
+    assert closes == [True]
+    assert "s1" not in mgr._stacks
+
+
+def test_cancelled_http_connect_closes_transport(monkeypatch):
+    """Deleting a server mid-connect cancels the connect task; the transport
+    must still be closed rather than leaked."""
+    async def initialize(self):
+        await asyncio.sleep(3600)
+
+    closes = _patch_http_transport(monkeypatch, initialize)
+    mgr = McpManager()
+
+    async def run():
+        await mgr._start_http_connect("s1", "remote", "http://x/mcp/", wait=0.05)
+        await mgr.disconnect_server("s1")
+        await asyncio.sleep(0)
+
+    asyncio.run(run())
+    assert closes == [True]

tests/test_model_discovery_model_ids.py

@@ -0,0 +1,33 @@
+"""A malformed provider row must not hide its valid neighboring models."""
+import pytest
+
+from src.model_discovery import ModelDiscovery
+
+
+@pytest.mark.parametrize('wrapped', [False, True], ids=['bare', 'openai'])
+@pytest.mark.parametrize('invalid_id', [None, '', 0, False, [], {}, 42, True, ['bad'], {'bad': 'id'}])
+def test_discovery_keeps_valid_models_beside_invalid_ids(monkeypatch, wrapped, invalid_id):
+    discovery = ModelDiscovery('fixture')
+    rows = [{'id': '/valid-model'}, {'id': invalid_id}, 'not a row', None, {'id': 'second-model'}]
+    payload = {'data': rows} if wrapped else rows
+
+    class Response:
+        is_success = True
+        def json(self):return payload
+
+    monkeypatch.setattr('src.model_discovery.httpx.get', lambda *_args, **_kwargs: Response())
+    monkeypatch.setattr(discovery, '_fingerprint_provider', lambda *_:'fixture')
+    result = discovery._check_port('fixture', 8080)
+    assert result is not None
+    assert result['models'] == ['/valid-model', 'second-model']
+    assert result['models_display'] == ['valid-model', 'second-model']
+    assert result['provider'] == 'fixture'
+
+
+@pytest.mark.parametrize('payload', [{'data': [{'id': 42}]}, [{'id': ['bad']}], {'data': []}])
+def test_discovery_with_no_valid_ids_stays_unavailable(monkeypatch, payload):
+    class Response:
+        is_success = True
+        def json(self):return payload
+    monkeypatch.setattr('src.model_discovery.httpx.get', lambda *_args, **_kwargs:Response())
+    assert ModelDiscovery('fixture')._check_port('fixture',8080) is None

tests/test_notes_clear_optional_fields.py

@@ -0,0 +1,82 @@
+"""Explicit null must clear nullable note fields; omission must preserve them."""
+
+import pytest
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from tests.helpers.database import disposable_database
+
+
+@pytest.fixture
+def client(tmp_path, monkeypatch):
+    import routes.note.note_routes as routes
+
+    with disposable_database(tmp_path) as factory:
+        monkeypatch.setattr(routes, "SessionLocal", factory)
+        app = FastAPI()
+
+        @app.middleware("http")
+        async def identity(request, call_next):
+            request.state.current_user = request.headers.get("x-test-user", "alice")
+            return await call_next(request)
+
+        monkeypatch.setattr(routes, "_scheduler_ref", None)
+        app.include_router(routes.setup_note_routes())
+        with TestClient(app) as api:
+            yield api
+
+
+@pytest.mark.parametrize("field,value", [
+    ("due_date", "2030-10-08T09:00:00"),
+    ("label", "home"),
+    ("image_url", "https://example.com/note.png"),
+    ("content", "Saved body"),
+    ("items", [{"text": "Saved item", "done": False}]),
+    ("color", "blue"),
+    ("gallery_id", "gallery-1"),
+    ("agent_session_id", "chat-1"),
+])
+def test_explicit_null_clears_saved_field(client, field, value):
+    created = client.post("/api/notes", json={"title": "Test note"})
+    assert created.status_code == 200, created.text
+    path = f"/api/notes/{created.json()['id']}"
+    populated = client.put(path, json={field: value})
+    assert populated.status_code == 200, populated.text
+    assert client.get(path).json()[field] == value
+
+    updated = client.put(path, json={field: None})
+    assert updated.status_code == 200, updated.text
+    saved = client.get(path)
+    assert saved.status_code == 200, saved.text
+    assert saved.json()[field] is None
+    assert updated.json()[field] is None
+
+
+def test_omitted_fields_survive_title_edit(client):
+    fields = {
+        "due_date": "2030-10-08T09:00:00",
+        "label": "home",
+        "image_url": "https://example.com/note.png",
+        "content": "Saved body",
+        "items": [{"text": "Saved item", "done": False}],
+        "color": "blue",
+        "gallery_id": "gallery-1",
+    }
+    created = client.post("/api/notes", json={"title": "Before", **fields})
+    assert created.status_code == 200, created.text
+    path = f"/api/notes/{created.json()['id']}"
+    updated = client.put(path, json={"title": "After"})
+    assert updated.status_code == 200, updated.text
+    saved = client.get(path).json()
+    assert saved["title"] == "After"
+    for field, value in fields.items():
+        assert saved[field] == value
+
+
+def test_null_update_cannot_clear_another_users_note(client):
+    created = client.post("/api/notes", json={"title": "Private", "label": "home"})
+    assert created.status_code == 200, created.text
+    path = f"/api/notes/{created.json()['id']}"
+    response = client.put(path, json={"label": None}, headers={"x-test-user": "bob"})
+    assert response.status_code == 404
+    assert client.get(path).json()["label"] == "home"

tests/test_notes_native_cleanup.py

@@ -0,0 +1,43 @@
+"""A detached Notes descendant must not hold output open or kill a bystander."""
+import os
+import runpy
+import subprocess
+import sys
+from pathlib import Path
+
+import pytest
+from src import process_lifecycle as lifecycle
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+@pytest.mark.skipif(os.name == "nt", reason="POSIX detached-session regression; Windows uses taskkill tree")
+def test_notes_cleanup_reaps_detached_pipe_holder_and_preserves_bystander():
+    stop = runpy.run_path(str(ROOT / "scripts/verify_notes_browser_runtime.py"))["_stop_owned_server"]
+    child = "import time; time.sleep(60)"
+    program = ("import subprocess,sys,time; "
+               f"p=subprocess.Popen([sys.executable,'-c',{child!r}],start_new_session=True); "
+               "print(p.pid,flush=True); time.sleep(60)")
+    with subprocess.Popen([sys.executable, "-c", child]) as bystander:
+        server = subprocess.Popen([sys.executable, "-c", program], stdout=subprocess.PIPE,
+                                  stderr=subprocess.STDOUT, text=True, start_new_session=True)
+        root_identity = lifecycle.ProcessIdentity.capture(server.pid)
+        detached = None
+        try:
+            child_pid = int(server.stdout.readline())
+            detached = lifecycle.ProcessIdentity.capture(child_pid)
+            assert os.getpgid(child_pid) == child_pid
+            stop(server, root_identity)
+            assert server.poll() is not None
+            assert detached.exited()
+            assert bystander.poll() is None, "cleanup touched an unrelated process"
+        finally:
+            # Test-owned fallback cleanup also runs when the negative control
+            # intentionally substitutes the old parent-only teardown.
+            targets = [identity for identity in (detached, root_identity) if identity]
+            lifecycle.terminate_identities(targets, steps=lifecycle.term_kill_steps(1))
+            server.wait(timeout=3)
+            if server.stdout:
+                server.stdout.close()
+            bystander.terminate()
+            bystander.wait(timeout=3)

tests/test_photo_editor_family.py

@@ -144,10 +144,12 @@ def test_editor_agent_endpoint_reports_scoped_tools():
 
 async def _route_contract(monkeypatch, message, *, attachment_meta=(), form=None, messages_out=None):
     from routes import chat_routes
-    from src import tool_security
+    from src import tool_security, tool_utils
     from tests.test_foreground_model_routing import _RouteRequest, _chat_stream_endpoint
 
     with monkeypatch.context() as route_patch:
+        # This built-in catalog fixture has no configured external MCP servers.
+        route_patch.setattr(tool_utils, "get_mcp_manager", lambda: None)
         endpoint = _chat_stream_endpoint(
             route_patch, "agent", {},
             session_model="odysseus-qwen3.5-tools-pre-heretic",