Follow-ups of the review of PR #418.
- check-latest-json.jq wants exactly one document and release.yml calls it
with -s: under jq 1.6, the release container's, an empty file passed the
check (no output, status 0) and two documents were judged on the last one,
so the job would have deleted the published manifest and uploaded the bad
one. Two tests cover both cases.
- build-latest-json.test.mjs runs the text of release.yml's "Publish
latest.json to package registry" step against a fake curl: DELETE then PUT
on a valid manifest, exit 1 and no request on a refused one.
- check-frontend no longer ignores .claude/** and release.yml (decision of
Max): the test reads verify-release.sh and runs the Publish step, so "on
every pull request" holds. REVIEW.md, CLAUDE.md and architecture.md follow.
- release.yml: stale ".deb fallback" comment fixed.
- qa-update-cycle.md: a tauri dev build on Linux shows TargetsNotFound,
expected since #418.
Refs #461Resolves#419
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
REVIEW.md, from the suggestions of the first review of #460:
- stacked-PR pitfall: fs_commands.rs compiles two CSV files of
src/__fixtures__/samples/ through include_str!, outside the paths of
check-rust;
- high level: .claude/settings.json and .mcp.json, a pull_request_target
trigger, an install step that changes the toolchain or the PATH (the
check-workflow list is now given as examples of the principle), a lock
origin whose host or scheme changes even with a new version, and a diff
that removes or bypasses a consent (price consent lives on the front);
- db.ts and dataExportService.ts stay normal (Max's decision of
2026-10-18), and the data zone now names the lock, the profile switch
and the restore;
- release zone: cargo-xwin and the Rust toolchain, installed unpinned at
each tag, also run with the signing key;
- windows-sys guard: EXPECTED joins CRATES and PIN;
- identifier: changing it orphans the databases and the app_data_dir
files (license key, activation token, token fallback); the keychain
service is a constant and does not follow.
scripts/review-levels.py is the calibration script of the PR body, now in
the repo: pages of 50 files read until an empty page (the instance caps
pages at 50), the new rules above, comment lines ignored by the workflow
checks, and --check, which compares its patterns with the Haut table
without network access. Fresh run: 3/28/8 on the 39 reference PRs and
6/33/12 on the 51, unchanged.
Refs #456
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
REVIEW.md sets the review level of a PR from the files its diff touches
(high, normal or low; the highest wins, never the author), lists what
matters in each zone with its source (a CLAUDE.md section or an ADR,
completed by the repo file where the fact lives), the verification
commands of the CI, and the review pitfalls of this repo. The pitfalls
taken from the global pr-review skill are corrected for this repo:
stacked PRs do get CI here since #232.
Calibration on the merged PRs #344 to #452: the 39 reference PRs of the
pilot document come out 3 low / 28 normal / 8 high (20.5 %, target
20 %), the whole range of 51 PRs 6 / 33 / 12 (23.5 %), against 3 / 19 /
17 with the trial patterns. A lockfile-only version bump of a package
already present stays normal, per Max's decision of 2026-10-17.
CLAUDE.md points to the file.
Fixes#456
Note: plan check bypassed after 3 passes, last findings applied (Max's
decision of 2026-10-18).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
