fix(meta): PR #175 review round — all 14 findings addressed

H1 godot-cold-parse exit-code guard (+bonus: import-pass for cold checkouts, found live); H2 pr-watchlist-diff loud registry failure; H3/T1 deny list :* normalization (add-only) + uniform allow syntax; H4/T7 helper-script allows; H5 get_api_key env-only (config.json is tracked — no secret fallback); H6 TEAM.md active/standby split; H7 troblum solo-profiling note; H8 pr-review frontmatter Task->Agent; H9 conventions doc taxonomies fixed vs real tree; T2 ask-gate leash files (settings/hooks) in tracked settings; T3 R-013 cross-reference; T4 governance README index regenerated (pql decisions sync); T5 dudley briefing ACTIVE; T6 pr-process step 7 run-from-main.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.claude/agents/README.md

@@ -107,7 +107,7 @@ Synthesize findings.
 ## Agent usage notes
 
 ### Troblum (Technical Consultant)
-- **Always paired with or supporting Tyre** - never works alone on architecture decisions
+- **Solo** for performance profiling, benchmarking, and technology stress-tests; **paired with Tyre** for architecture second opinions — never the sole voice on an architecture decision
 - Spawned for specific evaluation sidequests, not open-ended discussion
 - Blunt, data-driven, will challenge Tyre's assumptions with evidence
 

.claude/settings.json

@@ -54,8 +54,10 @@
 
       "Bash(godot4:*)",
       "Bash(gdformat:*)",
-      "Bash(tooling/atlas *)",
+      "Bash(tooling/atlas:*)",
       "Bash(tooling/validate-ron:*)",
+      "Bash(tooling/godot-cold-parse:*)",
+      "Bash(tooling/pr-watchlist-diff:*)",
 
       "Bash(chmod *)",
       "Bash(ls *)",
@@ -80,9 +82,32 @@
     ],
     "deny": [
       "Bash(git push --force *)",
+      "Bash(git push --force:*)",
+      "Bash(git push --force)",
       "Bash(git reset --hard *)",
+      "Bash(git reset --hard:*)",
+      "Bash(git reset --hard)",
       "Bash(git clean -f *)",
-      "Bash(rm -rf *)"
+      "Bash(git clean -f:*)",
+      "Bash(git clean -f)",
+      "Bash(rm -rf *)",
+      "Bash(rm -rf:*)"
+    ],
+    "ask": [
+      "Edit(.claude/settings.json)",
+      "Write(.claude/settings.json)",
+      "Edit(.claude/settings.local.json)",
+      "Write(.claude/settings.local.json)",
+      "Edit(.claude/hooks/**)",
+      "Write(.claude/hooks/**)",
+      "Edit(.config/hooks/**)",
+      "Write(.config/hooks/**)",
+      "Edit(.worktrees/**/.claude/settings.json)",
+      "Write(.worktrees/**/.claude/settings.json)",
+      "Edit(.worktrees/**/.claude/hooks/**)",
+      "Write(.worktrees/**/.claude/hooks/**)",
+      "Edit(.worktrees/**/.config/hooks/**)",
+      "Write(.worktrees/**/.config/hooks/**)"
     ]
   },
   "hooks": {

.claude/skills/pr-process/SKILL.md

@@ -285,9 +285,13 @@ git diff --stat main...<branch>
 Draft title (`<type>(<scope>): <summary>`, max 70 chars) and description.
 
 ```bash
-# NB: omit --repo — an explicit slug makes tea skip local-repo setup and
+# NB — both halves matter (.claude/rules/tea-cli.md):
-# pr create dies with "local repository required". Run from the MAIN checkout;
+#  1. omit --repo: an explicit slug makes tea skip local-repo setup and
-# tea infers the repo from origin. (.claude/rules/tea-cli.md)
+#     pr create dies with "local repository required";
+#  2. run THIS ONE COMMAND from the MAIN checkout, not this worktree —
+#     tea's go-git can't read a worktree's .git *file*. From the worktree:
+#     cd "$(git rev-parse --path-format=absolute --git-common-dir)/.."
+#     for this command, then return here for the remaining steps.
 tea pr create \
   --login schweitz \
   --title "<title>" \

.claude/skills/pr-review/SKILL.md

@@ -6,7 +6,7 @@ description: >
   /pr-review. Spawns reviewers matched to the branch type (code, copy, visual,
   audio) in parallel. Reports approve/reject with inline comments.
 user-invocable: true
-allowed-tools: Bash, Read, Grep, Glob, Task, Write
+allowed-tools: Bash, Read, Grep, Glob, Agent, Write
 ---
 
 # PR Review Skill

TEAM.md

@@ -31,16 +31,21 @@
 |-------|------|-------|-------------|
 | **SI** | Project Manager & Scrum Master | Sprint planning, ticket management, turning decisions into executable work. Manages `/ticket` skill. | Always (from now) |
 
+## Implementation Team (active)
+
+| Agent | Role | Focus | When active |
+|-------|------|-------|-------------|
+| **DUDLEY** | Server Developer | Game server, ECS, simulation loop, world state management. | Active — server/ is the main Phase-4 cascade workstream |
+| **STIG** | UI Developer | HUD, menus, insert/minimap, diegetic UI per D-013. | Active — client UI work as needed |
+| **JUSTINE** | Polish & Deploy | Build pipelines, packaging, performance optimization, release prep. | Active — build/validation tooling |
+| **MELLANIE** | Copywriter | In-game text, UI copy, tooltips, flavor text, lore entries. | Active — content batches |
+
 ## Standby Team (activate when needed)
 
 | Agent | Role | Focus | When active |
 |-------|------|-------|-------------|
-| **STIG** | UI Developer | HUD, menus, insert/minimap, diegetic UI per D-013. | UI implementation phase |
-| **DUDLEY** | Server Developer | Game server, ECS, simulation loop, world state management. | Server implementation phase |
 | **OSCAR** | Networking Developer | Multiplayer networking, client-server protocol, sync. | Networking implementation phase |
-| **JUSTINE** | Polish & Deploy | Build pipelines, packaging, performance optimization, release prep. | Pre-release phase |
 | **INIGO** | Sound Designer | Soundscape design, ambient layers, diegetic audio cues, audio propagation. | Audio implementation phase |
-| **MELLANIE** | Copywriter | In-game text, UI copy, tooltips, flavor text, lore entries. | Content creation phase |
 
 ## Agent Briefings
 

docs/briefings/dudley.md

@@ -1,6 +1,6 @@
 ---
 title: "Dudley - Project Briefing"
-description: "Backend/engine implementation agent on standby — covers Rust simulation server, ECS model, and bevy_ecs architecture decisions"
+description: "Server implementation agent — Rust/bevy_ecs simulation server, ECS model, world-generation cascade; the project's most active workstream"
 type: briefing
 status: active
 agent: Dudley
@@ -8,14 +8,19 @@ updated: 2026-03-13
 ---
 
 # Dudley - Project Briefing
-Last updated: 2026-03-13
+Last updated: 2026-07-13
 
 ## Current Project State
 Check active work: `pql ticket list --status in_progress`
 Check your tickets: `pql ticket list --assigned dudley`
 
 ## Status
-STANDBY. This briefing will be populated when backend/engine implementation begins. Dudley activates when simulation implementation work is assigned.
+ACTIVE. The Rust/bevy_ecs simulation server (`server/`) is the project's most
+active implementation workstream — the Phase-4 world-generation cascade (D-166,
+D-243) runs through it. Dudley is the default implementer for server batches
+selected by `/whats-next`; recent work includes the Layer-5 tile fill and the
+D-253 region transient model (PR #174). Build/test via the top-level Makefile;
+the pre-push gate runs fmt/clippy/full cargo test on every push.
 
 ## Key Decisions for Your Domain
 Read `decisions/architecture.md` (primary) and `decisions/content.md` (NPC model). Key decisions:

docs/design/diagram-wireframe-conventions.md

@@ -8,8 +8,9 @@ deleted (CLAUDE-CLEANUP §2 / T-1101).
 ## d2 diagrams (`/d2-diagram`)
 
 **Output taxonomy:** `.d2` source + `.png`/`.svg` renders under
-`docs/diagrams/{category}/{name}.d2` — categories in use: `architecture/`, `ui/`,
+`docs/diagrams/{category}/{name}.d2` — categories in use: `architecture/`,
-`entity/`. Create/update a diagram whenever a D-record is added or modified.
+`data-flow/`, `design/`, `entity/`, `state/`, `ui/`. Create/update a diagram
+whenever a D-record is added or modified.
 
 **Agent routing** (who to task with a diagram, by domain):
 - **Qatux** — architecture decision records, system-overview + data-flow diagrams.
@@ -19,8 +20,10 @@ deleted (CLAUDE-CLEANUP §2 / T-1101).
 ## Frame0 wireframes (`/frame0-wireframe`)
 
 **Output taxonomy:** wireframe JSON (source of truth) + exported PNG under
-`docs/design/wireframes/{category}/{name}.json` — categories include `insert/`
+`docs/design/wireframes/{category}/{name}.json` — categories:
-(neural-insert HUD), `dialogue/`, etc. Palette/style source is
+`character-creation/`, `dialogue/`, `hud/` (incl. the neural-insert HUD
+wireframes `v01-insert-basic.json`, `v10-insert-full.json`), `menus/`,
+`monologue/`, `navigator/`, `popups/`. Palette/style source is
 `docs/design/visual-grammar-v01.md`.
 
 **Agent routing:** Araminta (art direction / visual consistency) owns wireframe review;

governance/README.md

@@ -462,3 +462,4 @@ line in place — keep the Q-record for the audit trail rather than deleting it.
 - [R-010: protobuf for client-server serialization](rejected/architecture.md#r-010-protobuf-for-client-server-serialization) — _architecture_
 - [R-011: Single currency for Phase 2 economics](rejected/economics.md#r-011-single-currency-for-phase-2-economics) — _economics_
 - [R-012: Overheard NPC conversation system (D-078) — scrapped](rejected/perception.md#r-012-overheard-npc-conversation-system-d-078--scrapped) — _perception_
+- [R-013: Localization / i18n](rejected/scope.md#r-013-localization--i18n) — _scope_

governance/rejected/scope.md

@@ -4,4 +4,5 @@ Rejected proposals in the **scope** domain, rationale preserved for the audit tr
 
 ### R-013: Localization / i18n
 - **Rejected:** 2026-07-13
+- **Cross-reference:** D-123 (content.md — Generative AI for NPC content), D-138 (content.md — LLM re-voicing pipeline); roster removal in `.claude/agents/README.md` + `TEAM.md`
 - **Reason:** Full localization — an i18n framework, translation management, and the volume of template writing the Settled Reach's procedurally-composed text would demand — is prohibitively expensive for a solo developer, and compounds against the Gemma-based naming/text generation: every generated string would need a parallel translation pathway. The game ships English-only. The `tiger` (Translator) agent persona was removed with this decision (see `.claude/agents/README.md` and TEAM.md). Revisit only if project scope and resourcing change fundamentally.

tooling/db/common.py

@@ -75,24 +75,25 @@ def get_base_url(key: str, default: str) -> str:
     return load_config().get(key, default)
 
 
-def get_api_key(env_var: str, config_key: str) -> str:
+def get_api_key(env_var: str, config_key: str = "") -> str:
-    """Get an API key from the environment or config.json.
+    """Get an API key from the environment — environment-only, by design.
 
-    Checks the ``env_var`` environment variable first, then ``config_key`` in
+    ``tooling/db/config.json`` is a *tracked* file and holds endpoints only;
-    config.json. Prints a JSON error and exits 1 if neither is set — connector
+    it must never carry secrets, so there is deliberately no config.json
-    scripts emit machine-readable JSON on all paths.
+    fallback here (the old one steered users toward committing paid API keys).
+    ``config_key`` is retained in the signature for caller compatibility but
+    is ignored. Prints a JSON error and exits 1 if the variable is unset —
+    connector scripts emit machine-readable JSON on all paths.
     """
     key = os.environ.get(env_var)
     if key:
         return key
-    try:
-        with open(CONFIG_PATH) as f:
-            config = json.load(f)
-            return config.get(config_key, "")
-    except Exception:
-        pass
     print(json.dumps({
         "ok": False,
-        "error": f"No {env_var} found in environment or config.json"
+        "error": (
+            f"{env_var} not set. Export it in your shell or add it to the "
+            "machine-local .claude/settings.local.json env block (untracked). "
+            "Never put keys in tooling/db/config.json — it is tracked."
+        )
     }, indent=2))
     sys.exit(1)

tooling/godot-cold-parse

@@ -28,15 +28,50 @@ RUN_MENU=false
 
 rm -f "$REPO_ROOT/client/.godot/global_script_class_cache.cfg"
 
+# A truly cold checkout (fresh clone or worktree — .godot/ is gitignored) has
+# no resource-import cache, and every imported asset (fonts, ogg) then "fails
+# loading" during the parse run: a wall of false positives. Seed the cache
+# with an import pass first; source assets are tracked, so this is always
+# reconstructible. (Found live: first run in a fresh worktree, 2026-07-13.)
+if [ ! -d "$REPO_ROOT/client/.godot/imported" ] || [ -z "$(ls -A "$REPO_ROOT/client/.godot/imported" 2>/dev/null)" ]; then
+    echo "godot-cold-parse: no import cache — running one-time import pass..." >&2
+    set +e
+    IMPORT_OUT=$(godot --headless --path "$REPO_ROOT/client" --import 2>&1)
+    IMPORT_EXIT=$?
+    set -e
+    if [ "$IMPORT_EXIT" -ne 0 ]; then
+        echo "godot-cold-parse: import pass exited $IMPORT_EXIT" >&2
+        printf '%s\n' "$IMPORT_OUT" | tail -20 >&2
+        exit "$IMPORT_EXIT"
+    fi
+fi
+
 FILTER='^(SCRIPT )?ERROR|Parse Error|Export type'
 
-MATCHES=$(godot --headless --path "$REPO_ROOT/client" --quit 2>&1 \
+# Capture the godot run separately from the filter pipeline: with the
+# trailing `|| true` on the greps, a nonzero exit from godot itself (crash,
+# missing binary, corrupted install) would otherwise report "clean". Nothing
+# downstream re-reads the raw output now that this is scripted, so fail loud.
+set +e
+RAW=$(godot --headless --path "$REPO_ROOT/client" --quit 2>&1)
+GODOT_EXIT=$?
+set -e
+if [ "$GODOT_EXIT" -ne 0 ]; then
+    echo "godot-cold-parse: godot itself exited $GODOT_EXIT — not a parse verdict" >&2
+    printf '%s\n' "$RAW" | tail -20 >&2
+    exit "$GODOT_EXIT"
+fi
+
+MATCHES=$(printf '%s\n' "$RAW" \
   | grep -iE "$FILTER" \
   | grep -v "Failed loading resource: res://assets" \
   | grep -v "Cannot infer the type" \
   | grep -vE '(Messagepack|LocalBridge|ServerProcess|Constants)" not declared' || true)
 
 if [ "$RUN_MENU" = true ]; then
+    # Deliberately no exit-code check here: `timeout` kills the menu after
+    # 10s by design (exit 124 is the expected shutdown path); only the
+    # scraped error lines carry signal for this bounded run.
     MENU_MATCHES=$(timeout 10 godot --path "$REPO_ROOT/client" res://scenes/main_menu.tscn 2>&1 \
       | grep -iE "$FILTER" || true)
     if [ -n "$MENU_MATCHES" ]; then

tooling/pr-watchlist-diff

@@ -17,7 +17,16 @@ set -euo pipefail
 BASE="${1:?usage: tooling/pr-watchlist-diff <base> <head>}"
 HEAD="${2:?usage: tooling/pr-watchlist-diff <base> <head>}"
 
-mapfile -t GENERATOR_SOURCES < <(python3 tooling/generator_sources.py --list)
+# Load the registry via plain assignment (set -e sees its failure), not
+# `mapfile < <(...)` — a failed process substitution is invisible to set -e
+# and would silently yield an empty watch list, disabling the DB-staleness
+# net exactly when the shared registry breaks. Guard the empty case too.
+SOURCES_RAW="$(python3 tooling/generator_sources.py --list)"
+if [ -z "$SOURCES_RAW" ]; then
+    echo "pr-watchlist-diff: generator_sources.py --list returned nothing" >&2
+    exit 1
+fi
+mapfile -t GENERATOR_SOURCES <<< "$SOURCES_RAW"
 
 git diff --name-only "$BASE...$HEAD" -- \
   "${GENERATOR_SOURCES[@]}" \