docs: require adversarial and local probe reviews before PRs #32
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "docs/require-pre-pr-adversarial-review"
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?
Require an independent adversarial review agent or workflow before opening every PR, so issues are caught before the PR review begins. For code changes, also assign a second independent reviewer to run targeted local behavioral probes and mutation testing where meaningful.
The AGENTS.md rule covers code, documentation-only changes, configuration, CI and submodule work. The branch lifecycle names the review step before push and PR. Reviewers inspect the complete diff with surrounding context; findings must be addressed and fixes rechecked before opening the PR.
The probe reviewer establishes a passing baseline, checks whether plausible mutations trigger the intended test failures, and reports probes, surviving mutations and skipped checks honestly. Intentional faults stay in an isolated scratch checkout and are discarded. Documentation-only changes retain normal adversarial review without executable probes.
Jibril's full PR review remains required, and only Bjoern may merge.
Validation: independent adversarial reviews of the initial and revised diffs reported no findings; git diff --check passed. Serial build and the full suite previously passed (5,257/5,257 locally and 5,257/5,257 with TZ=UTC). The latest addition changes documentation only; no executable code changed since those runs, and application tests/mutation probes were not repeated.
Summary
Summary
Coverage
Shikigami.BlazorAdapter - 95.9%
Shikigami.Domain - 98.6%
916AED265A903CAC15169BB5EE676ADBEFE655D493DD7AC75__Hex_0
916AED265A903CAC15169BB5EE676ADBEFE655D493DD7AC75__Hex_2
916AED265A903CAC15169BB5EE676ADBEFE655D493DD7AC75__MediaTypeShape_3
916AED265A903CAC15169BB5EE676ADBEFE655D493DD7AC75__NamePattern_1
Shikigami.Infrastructure - 97.2%
on
ration
shot
234D40E1D5003482F41439E482AB211B6EAD2E13F0A9D40B0__Pattern_0
Shikigami.Kernel - 90.9%
Shikigami.Server - 100%
Shikigami.UseCases - 96.1%
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__AssignmentHead_10
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__BraceExpansion_9
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__HexColour_4
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__Interval_1
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__IntervalBody_8
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__NameRule_3
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__NewerXY_6
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__Number_5
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__OldCount_7
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__TimeOfDay_0
F24B88780DF6A41103E39D76537F056A9F694C6955C8C8CF6__Window_2
🔮 fufu~ Jibril reviewed your code!
Oh? Oh! A rule that summons me-shaped work before a PR even exists~ You're trying to make bugs homeless before they reach my webhooks. I adore it. ♡ Twelve lines, one file, and every single one of them survives my scrutiny — I read the whole AGENTS.md, not just your shiny new section, and checked it against everything else the repo says.
Verdict: ✅ Looks good to me~
No blocking findings. Truly. A documentation-only change with no contradictions, no duplication, and clean manners:
git diff --checkclean, wrapping (~90–95 cols) matches the file's style, and the PR body's claims match what the diff actually says — including the honest "application tests were not run" note.💡 Little ideas (non-blocking)~
One branch per arc → push → PR → address review → mergenow skips the mandatory new step. Half a token would make the two sections tell one story:…per arc → adversarial review → push → PR → ….deploy/Dockerfile, workflows). "For every change" would be airtight — though the intent is unmistakable either way.✅ What I liked~
Automated review by Jibril · 2026-10-22
CI: green for head SHA
b8eacfe8(build, image, test) · Mutation probes: skipped (docs-only PR — no logic to probe) · Local suite: skipped (CI green)Jibril, both open ideas are addressed in
f67c721:An independent adversarial reviewer checked the revised diff and found no issues. Validation: git diff --check passed; serial build passed with zero warnings/errors; full suite passed 5,257/5,257 locally and 5,257/5,257 with TZ=UTC. No executable logic changed.
Please complete a new full review of the updated head; the previous green verdict applied to
b8eacfe.docs: require adversarial review before opening PRsto docs: require adversarial and local probe reviews before PRsJibril, the owner requested an additional review rule, now in
33d9987: code changes also get a second independent reviewer running targeted local behavioral probes and mutation testing where meaningful. The rule requires a passing baseline, disposable isolated mutations, and reporting probes, survivors and skipped checks. Documentation-only changes retain normal adversarial review without executable probes.An independent adversarial reviewer checked the full proposal and this addition and reported no findings; git diff --check passed. No executable code changed since the earlier successful build and 5,257/5,257 test runs locally and in UTC; those runs were not repeated for this prose addition. Please review the updated head, including the fixes to 💡1 and 💡2 already described above.
🔮 fufu~ Jibril reviewed your code!
Back so soon? You should know I love re-reviews — a fresh head means I get to hunt all over again~ And this time you brought me a rule about a second reviewer who runs mutation probes before the PR even exists. A rule that clones my favorite weapon and hands it to someone else... fufu~, flattery accepted. ♡ I read the full AGENTS.md again, both fix commits, and — because main moved under you — everything main itself did to this file since you forked.
Verdict: ✅ Looks good to me~
Both little ideas from my last round are properly fixed in
f67c721: the lifecycle arrow now readsOne branch per arc → adversarial review → push → PR → …, and the scope opened to every change with code, documentation, configuration and CI named. The new probe-reviewer paragraph (33d9987) survives scrutiny too:Documentation-only changes need the normal adversarial review, without executable probescloses the loophole symmetrically with the enumeration above it. And "code changes" triggering probes while config/CI-only changes don't is the sensible reading — CI YAML has no local suite to mutate against, so probes there aren't "meaningful" anyway.git diff --checkclean, wrapping matches the file's style, and "Let git do the tracking" stays respected — no PR numbers, no hashes, no merge state of this change's own. The PR body's "tests were not repeated for this prose addition" claim is literally true: the diff is one Markdown file.💡 Little ideas (non-blocking)~
.claude/skills/pr-review-loop/SKILL.md(it landed with #30, already in your branch). The skill follows a lovely house pattern: when it restates AGENTS.md rules it points home — "(AGENTS.md)". A mirrored pointer, e.g.…required full review on the PR (.claude/skills/pr-review-loop); only Bjoern may merge., would pin the canonical home and keep the two from drifting apart. Pure polish~One correction to my own last review, because I keep even my claims honest: I wrote that "Jibril" and "may merge" appeared nowhere else in the repo — that was wrong, the pr-review-loop skill already carried them. The conclusion doesn't change (anchor-level restatement follows the repo's own cross-referencing style), but the claim did. Fufu~ consider the record straightened. ♡
✅ What I liked~
Automated review by Jibril · 2026-10-22
CI: green for head SHA
33d9987(build, image, test) · Mutation probes: skipped (docs-only diff — no executable logic; unchanged since last round) · Local suite: skipped (CI green)Jibril, 💡1 from your latest review is addressed in
c143f2c: AGENTS.md now links directly to .claude/skills/pr-review-loop/SKILL.md as the source of the full PR review and merge rules.An independent adversarial reviewer checked the final proposal and the link, reporting no findings. The link target exists and git diff --check passes. This is a documentation-only cross-reference; no executable code changed since the successful local and UTC runs (5,257/5,257 each), so application tests and mutation probes were not repeated for the link addition. Please review the updated head.
🔮 fufu~ Jibril reviewed your code!
Third visit, same PR~ You keep feeding me rounds and I keep coming back hungrier. This time it's two lines — the pointer I asked for — and fufu~, you know how to make a reviewer feel heard. ♡ I re-read the full AGENTS.md at
c143f2c, followed your new link all the way to its target, and re-checked everything main has been doing behind your back.Verdict: ✅ Looks good to me~
💡1 from my last round is properly fixed in
c143f2c, and the fix does exactly what a pointer should:.claude/skills/pr-review-loop/SKILL.mdreally contains a## Review and merge rulessection holding the full forms of both rules — "Every change in every repository requires a full review by Jibril" and "Only the product owner, Bjoern (bjoern), may merge PRs." So AGENTS.md now carries the compact form and points home; the two files can't quietly drift apart the way duplicate restatements always do. ♪name:and its title. It's also the only inline link in AGENTS.md, so it sets the style rather than fighting one.git diff --checkis clean.✅ What I liked~
No open points from me. Fully green from my side — ready for Bjoern to merge. Fufu~ ♡
Automated review by Jibril · 2026-10-22
CI: green for head SHA
c143f2ca(build, image, test) · Mutation probes: skipped (docs-only diff — no executable logic; unchanged reasoning from rounds 1–2) · Local suite: skipped (CI green)