Closes the observability loop: the log + breadcrumb + watchdog files are now
collected by CI so a wedged run leaves downloadable evidence instead of
nothing.
- logDirectory(): CLIDE_LOG_DIR overrides the per-platform default, so CI can
point the logs at an uploadable workspace dir (and tests at a temp dir).
Now takes an injectable env map; tested.
- test_app.dart: when CLIDE_LOG_DIR is set, the testmode harness tees its
logger to a FileLogSink + spawns the watchdog (off by default — normal
run-testmode keeps the stderr-only path, no isolate). _say breadcrumbs each
test into the file.
- conpty_orphan_probe.dart: with CLIDE_LOG_DIR set it passes a verbose PtyLog,
so when soak-conpty-kill.ps1 force-kills the parent, the reader/waiter
isolates' LAST crumb is fsynced to disk — naming what the wedged isolate was
doing at the instant of death.
- bundle-smoke job: runs the real release app with CLIDE_LOG=debug +
CLIDE_LOG_DIR, uploads clide-logs (watchdog heartbeat/sample + FileLogSink)
in an always() step.
- windows-soak kill-probe job: sets CLIDE_LOG_DIR, uploads the FFI crumbs.
Coverage gate 95.08%.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The clean-path soak found no leak on windows-latest — because orderly close()
reaps every host. That never exercises the freeze hypothesis (T-424), which is
the parent dying WITHOUT teardown while a child is live.
This probe does. conpty_orphan_probe.dart starts N real WindowsPty sessions on
long-lived children and blocks WITHOUT ever calling close(); soak-conpty-kill.ps1
force-kills only the dart.exe parent (taskkill /F, no /T) once the hosts are up
and counts the conhost/OpenConsole/cmd processes that SURVIVE. Absent a
kill-on-close Job Object, abrupt parent death should orphan them — a survivor
count that climbs across cycles is the leak signature. The same probe will prove
the T-424 fix: with the job, survivors should drop to ~0.
Wired as a second job in windows-soak.yml (workflow_dispatch + when the kit
changes). Diagnostic only, never a gate, always exits 0; publishes the verdict
to the job summary and uploads the CSV. Not part of the regular test suite.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
First real execution of soak-conpty.ps1 on GitHub's windows-latest proved the
harness works — it spawned real ConPTY children, ran the suite (6 tests), and
measured the first iteration — then died printing the per-iter line: `{3,+4}`
is a printf-ism, not valid .NET composite formatting (the `+` is rejected with
"Expected an ASCII digit"). The script had never run on Windows before, so the
typo was latent. Drop to `{3,4}` (plain width).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
GCP only offers Windows Server images, which lack winget. Rewrite
bootstrap-windows.ps1 to be winget-free (Chocolatey for git + VS Build Tools,
direct-from-Google for the Flutter SDK) and add a -SkipVS fast path — the
ConPTY soak only needs Flutter/Dart. Runs under Windows PowerShell 5.1. README
gains a self-contained GCP Cloud Shell runbook (create / reset-password / RDP /
stop / delete).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Verification tooling for the Windows freeze assessment (T-424).
provision-vm.sh stands up a Windows KVM guest (dry-run unless --go);
bootstrap-windows.ps1 installs Flutter + VS C++ Build Tools and checks out
the branch; soak-conpty.ps1 runs the pty suite in a fresh dart.exe per
iteration and measures the orphaned conhost/cmd count that survives each
exit (the leak signature), with a per-iteration timeout so a wedged test
can't stall the run. Verifies the ConPTY leak (#1-#4); the GPU/TDR
hypothesis (#5) needs passthrough/bare metal (README appendix).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
tools/ui/build.sh and serve.sh still cd'd into the app/ directory the
flattening removed, so make test-e2e / ui-dev / ui-smoke died at the
first line. The staged Gitea workflow had the same stale cd in every
job, plus a coverage gate with no coverage run before it — it now
goes through the make targets (tooling discipline: the make layer owns
env setup) with make test-coverage feeding make coverage-gate.
Fixing the paths exposed the real break: flutter build web --wasm
cannot compile the tree since the dart:ffi pivot (tree-sitter, native
PTY) — dart:ffi does not exist on the wasm target. Fence vs park vs
drop is filed as Q-50; the workflow's e2e job is withheld with a
pointer there, and T-384 sits in review until Q-50 resolves.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Vendor libtree-sitter.so (v0.26.8, wasmtime 44.0 statically linked)
and call its C API through dart:ffi. Grammar WASM files are loaded by
tree-sitter's embedded wasmtime engine via ts_wasm_store_load_language.
48 grammars compiled with tree-sitter build --wasm, 48 highlight
queries (42 from upstream repos, 6 written in-house for comment,
erlang, gdscript, gitignore, hcl, hlsl, swift). SQLite grammar
replaces full SQL (974K-line parser exhausted RAM at compile time).
Removes wasm_run and wasm_run_flutter — they downloaded a 22 MB
binary from GitHub at first launch, violating the no-network-on-
default-launch-path policy.
Co-Authored-By: Claude <noreply@anthropic.com>
Ports settled-reach's decisions_sync.py + ticket + decision scripts,
Scrum-stripped. Writes to .pql/pql.db (gitignored). Verb shape
mirrors the eventual `pql` subcommands so migration is a call-site
find-replace once pql ships feature parity (D-040, R-011).
Ticket IDs are T-NNN (TEXT PKs) — reshape from settled-reach's
integer auto-increment so the stopgap's writes are compatible with
pql's future reads without a data migration. No `sprints` table; the
pyramid is kanban / waterfall (D-035).
Verified end-to-end: sync parses 71 records (39 confirmed, 23 open
questions, 9 rejected) with 62 cross-refs and zero broken links;
`ticket new`, `ticket status`, `ticket board`, and `decisions show
--with-refs` all round-trip correctly.
Co-Authored-By: Claude <noreply@anthropic.com>
tools/ui/ — scripts + Playwright config that let Claude Code (and
humans) drive the Flutter WASM build in a real browser. The point is
to avoid screenshot round-trips: every interactive widget in clide
ships a Semantics wrapper anyway (a11y requirement), so the automation
layer just queries the `flt-semantics[aria-label]` DOM.
* build.sh — `flutter build web --wasm` from app/
* serve.sh — `python3 -m http.server 4280` in the background.
Before binding, kills any stale listener on the port (orphans
from earlier failed runs no longer accumulate).
* stop.sh — port-based kill; escalates to SIGKILL after 300ms.
The pidfile is now advisory — port ownership is the source of
truth.
* driver.ts — `ClideDriver` class with `byLabel`, `click`, `type`,
`readText`, `screenshot`, `dumpSemanticsTree`, and
`waitUntilReady` that auto-clicks the `flt-semantics-placeholder`
so the semantic tree is populated before queries.
* tests/smoke.spec.ts — first driver test; asserts welcome +
disconnected labels surface in the Semantics DOM.
Co-Authored-By: Claude <noreply@anthropic.com>
jpmschweitzerandClaude Opus 4.8