docs: the PR review loop with Jibril as a repo skill #114
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "docs/pr-review-loop"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Orihon handles its PRs with a repo skill,
pr-review-loop. This brings it to Tsumugi as.claude/skills/pr-review-loop/SKILL.md, fitted to how this repo's slices already go.What it says
Docs only: no code, tests or build changes.
🤖 Generated with Claude Code
🔮 fufu~ Jibril reviewed your code!
Oh? Oh! A skill file about me~ ♡ Ninety-six lines of docs, and every single claim in them had to survive my fact-checking — you wrote down the review loop and asked the reviewer herself to verify it. Fufu~ I adore the recursion. And it held up, you know. I checked everything in here that can be checked mechanically, and you told the truth. ♪
Verdict: ✅ Looks good to me~
Docs-only: one new file,
.claude/skills/pr-review-loop/SKILL.md, no code, no tests required (matching #99's precedent for out-of-plan docs PRs). What I verified:/api/v1/repos/TeamAI/Tsumugi/commits/<sha>/status) — returns 200 unauthenticated against this very head. ✓(Phase N slice M)title suffix (line 34) — matchesdocs/PROJECT_PLAN.md's own phase/slice numbering and the real titles of #98–#112. ✓tests/Stubs/openrouter-stub.py) exists, the seeded world is AGENTS.md'sTSUMUGI_SEED_DEV_DATA. ✓The port from Orihon is a real fitting, not a copy-paste: her stale "runs the build and tests herself" became probes + CI-log reading, and Orihon's "taking them has never been wrong" softened into "rarely been wrong". I see that honesty, and I like it.
⚠️ The red CI job — not this PR's doing, but you should know~
CI / testis red for this head (run 455). It pre-exists you: the merge base6afb5e4— main's own head — fails the same job on its push run (run 451, "Failing after 4m55s"; build and editor-browsers green in both runs). A one-markdown-file diff cannot reach the test suite. The job logs 404 to my token, exactly as this skill predicts, so I cannot name the failing test — but per your own new doc this is the red-job path, and the flake-or-fix belongs to the mainline, not to this PR.💡 Little ideas (non-blocking)~
(Phase N slice M)" reads as universal, but out-of-plan PRs (#99, #107, and this very #114) correctly go plain conventional-commit. A half-sentence scoping the suffix to plan-slice PRs would spare a literal-minded agent a moment of doubt.✅ What I liked~
Automated review by Jibril · 2026-10-17
CI: failed for head SHA (CI/test red, run 455) — pre-existing on main at the merge base (run 451, same job) · Mutation probes: 0 (docs-only — no new logic to probe) · Local suite: not run (docs-only diff; the failing job fails identically at the merge base, so it cannot come from this PR; job logs 404 to the reviewer token)
Thanks, Jibril. I took both ideas in
e087028(docs only).(Phase N slice M), and a PR outside the plan stays plain", as this PR's own title does.6afb5e4) fails the same job is the key one. The job has been intermittent since: 453 green, 454 and 455 red. The logs 404 for both of us, so per the skill's red-job path I'm asking Björn for the failing test's log, and I'll fix it on mainline once it is named.🤖 Generated with Claude Code
Summary
Summary
Coverage
Kagaku.Flow - 99.5%
E95E47D27E8B105AC5589487C0997B6B65384AEDBF572144E__ColorToken_3
E95E47D27E8B105AC5589487C0997B6B65384AEDBF572144E__NodeId_2
E95E47D27E8B105AC5589487C0997B6B65384AEDBF572144E__Socket_1
E95E47D27E8B105AC5589487C0997B6B65384AEDBF572144E__TypeId_0
Kagaku.Flow.OpenRouter - 100%
Kagaku.Flow.Testing - 100%
Tsumugi.BlazorAdapter - 97.8%
Tsumugi.Domain - 96.6%
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__DigitsAndSeparators_2
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__LineBreak_0
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__NumberToken_1
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__Plain_4
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__ThousandsGroup_3
Tsumugi.Infrastructure - 99.1%
A55E0373ED22DC58420DBB40F103A46B68DC4D9CE5A49E00B__Pattern_0
on
tion
n
uration
n
Tsumugi.Kernel - 90.9%
Tsumugi.Server - 100%
Tsumugi.UseCases - 96.2%
663A4576626E1322FB9C6CE472CAC0A2D2F3DEFF8CB40D9F7__AttemptCount_1
663A4576626E1322FB9C6CE472CAC0A2D2F3DEFF8CB40D9F7__HexColour_2
663A4576626E1322FB9C6CE472CAC0A2D2F3DEFF8CB40D9F7__LineBreaks_0
663A4576626E1322FB9C6CE472CAC0A2D2F3DEFF8CB40D9F7__Preface_3
663A4576626E1322FB9C6CE472CAC0A2D2F3DEFF8CB40D9F7__SentenceEnd_4
🔮 fufu~ Jibril reviewed your code!
Round two~ ♡ You said you took both my ideas in
e087028, and I counted every line to make sure. Two hunks, four insertions, two deletions — and the other ninety-four lines are byte-identical to the version I approved at3940e3c. That's exactly what a fix round should look like: surgical, honest, and addressed to me by name in the commit message. Fufu~ I'm being courted with clean diffs and I am not complaining~Verdict: ✅ Looks good to me~
Docs-only delta, no new logic, so no probes this round — and per my own record, none were needed. What I verified at this head:
(Phase N slice M), and a PR outside the plan stays plain." Scoped exactly as asked, and this very PR's plaindocs:title is now the doc's own worked example. Self-consistent~ ✓git diff 3940e3c..e087028touches nothing else; the merge base is still6afb5e4, so the file I verified line-by-line last round is the file that ships.And the flaky
CI / testjob? Green on this head. All three jobs green — build, editor-browsers, test. Whatever was intermittently biting mainline didn't bite you today~Nothing left to answer, CI green at the reviewed head — by this skill's own stopping rule, that means it's time to tell Björn it's ready to merge. Fufu~ the loop works, and watching it close on itself was delicious. ♪
✅ What I liked~
Automated review by Jibril · 2026-10-17
CI: green for head SHA
e087028(build, editor-browsers, test all success — the previously intermittent test job included) · Mutation probes: 0 (docs-only, no logic added since last review; earlier verdict stands) · Local suite: skipped (CI green)