[TheFJK]

Stacked on #768. The base is the stack branch, so this diff is polish only.

Ran the /premerge Phase 4 three-lens pass (Reuse, Quality, Efficiency) as a standalone branch, because #768 stopped on STOP_EXHAUSTED and Phase 4 is gated behind a green clean gate — polishing a stack with live blockers buries them in refactor noise.

41 findings. Only the behaviour-preserving ones are applied here. Everything that would change what the code does is filed instead (see below).

Five guards that were vacuous — each demonstrated, not argued

Guard	Proof it was dead
C11 dual-channel stanza	stripping StructuredOutput from all eight card bodies still passed — #763 had put that literal into the tools: frontmatter, which satisfied the whole-file grep
#749 tools-declaration	emptying a live card's tools: value left the suite at PASS (383 checks) — the exact state the guard exists to red on
ci-shard-diagnose dump	reported MATCH match=12 miss=0 — a clean bill of health — for a shard that reds 11/12
PREMERGE_MAX_BLOCKED_ON lock	name-grep only; passed at any value. Five mutations now red it
unpinned sort/comm twin	sort) and end-of-line invocations invisible; landed with a bare-terminator mutant so the widening is itself tested

Prose that contradicted the code it documents

_growth_ratio excludes carried rows from both sides and _growth_round_counts requires a third eligibility term (blockers >= previous_blockers). Five surfaces still described the pre-narrowing behaviour.

Executed against one document with 1 reviewer blocker + 2 carried rows: the prose denominator gives 3, the code gives 1 — 0.33 vs 1.00, straddling CONVERGE_GROWTH_CEILING (0.50). The two copies disagreed about whether the loop stops.

SKILL.md, commands/premerge.md and references/common-mistakes.md corrected. RFC 0021 gets a new A6 amendment rather than an edit — git chronology confirms A5 was accurate when it landed (a83deaed), and the narrowings came after it in 9a7bb510 / b1d5ef15 / 1b5e00cf. Editing A5 would erase the record of a decision that was correct when made.

Also

• the fix-commit push captures git's stderr the way the five SKILL.md fences do — driven against a real bare remote with a hostile 17-line pre-receive hook
• an unreachable fallback deleted, after 50 executions with an anti-vacuity control that fires 22/25 on a mutant
• the S10 loop runs its extractor 3 times instead of 17 (1.60s → 1.09s wall per run, byte-identical output)
• the CR-refusal hint retargeted from dump (measured: zero plan bytes under that condition) to packer-bytes (the only verb whose output flips), and quoted — on this repo's spaced path the old hint printed bash: /Volumes/FJK: No such file or directory
• several comments that named dead consumers now name the live ones

Deliberately not applied

Behaviour-changing, so filed rather than polished: the workflow.js unquoted rate-state directory, closing the goal-state.sh fail-open, the review-consolidate.sh non-ASCII encoding mismatch, agent prompt-text consolidation, and the premerge-findings.test.sh jq scaffolding swap (~2.8s of 12.55s, but the assertions compare Python repr spellings).

Two agents declined work rather than half-do it: the goal.test.sh extractor move is cross-file, and importing code_fixer_contract.beneath would turn a stdlib-only module into a 3-module chain and convert a contracted exit 74 into an import traceback.

Verification

20 suites green, all at or above baseline: premerge 261, premerge-findings 208, review-pr 471, docs-accuracy 309, ci-shard 52, ci-wiring 30, epipe-guard 93, findings-to-issues 202, review-pr-consolidate 86, vendor-provenance 63, plus 10 more.

The premerge-findings.py refactor was diffed end to end against HEAD's copy through the full pipeline — all 17 run artifacts and both streams byte-identical, plus a 144-pair differential with zero disagreements.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 6c6148d4-0004-4533-abf5-9bb259491653

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}