Skip to content

moonlight-lupin/agent-skills / plugins/dev-pipeline/skills/review-commit

review-commit

Verified

Use when the user says 'review loop N times', 'build then review', 'fix and re-review', 'review loop until clean', 'have cursor build, codex review', or wants the dev-pipeline review stage: a two-axis single pass (Standards + Spec, parallel, never merged) FIRST, then a budgeted confirm-round loop on the fix diffs until no BLOCKING/MAJOR is open, then the commit/deploy gate — three-reviewer gate before live deploy, release hygiene before tagging. Also triggers: 'two-axis review', 'confirm rounds', 'review diff until SHIP', 'N confirm rounds', ship gate.

FileKindSize
SKILL.mdexecutableskill md6.8 KB
references/background-build-polling.mdexecutablescript3.9 KB
references/confirm-round-protocol.mdexecutablescript12.3 KB
references/dispatch-result-contract.mdexecutablescript3.5 KB
references/feature-build-brief.mdexecutablescript2.9 KB
references/fix-brief-template.mdexecutablescript10.5 KB
references/persistent-loop-plan.mdexecutablescript6.9 KB
references/release-hygiene.mdexecutablescript3.9 KB
references/reviewer-dispatch.mdexecutablescript6.0 KB
references/spec-convergence.mdexecutablescript5.2 KB
references/two-axis-single-pass.mdexecutablescript9.9 KB
references/workflow-steps.mdexecutablescript18.9 KB
scripts/attest_file.pyexecutablescript2.7 KB
scripts/capture_dispatch_result.pyexecutablescript3.6 KB
scripts/loop_plan_check.pyexecutablescript3.8 KB

references/spec-convergence.md

View on GitHub
# Spec Review Convergence

Multi-round review of a SPEC (or plan/design doc) until a SHIP verdict, before any implementation. Distinct from code-diff confirm rounds and the `test-evals` batched per-batch review: the artifact under review here is the document, and the quality gate is reviewer convergence.

## Round 1: two diverse reviewers

Dispatch at least two reviewers with different strengths on the SAME axes. Proven pair: reviewer lanes A (Codex) + B (Claude Code Opus) — models in `orchestrator/references/lanes.md`. They catch different classes:
- Codex: completeness, contradictions, scope risk, feasibility against the actual codebase (it cites file:line when given repo access via `codex exec --sandbox read-only -C <repo>`).
- Claude Code Opus: invariant conflicts, untestable acceptance criteria, and mechanisms the codebase doesn't actually provide. Use `--permission-mode plan`; `--max-turns 25` if it must read files (each Read = 1 turn; name exactly which files).
- Reviewer lane C (Muse) is a strong third eye — see `muse-review-quirks` for invocation. It excels at re-observation/race gaps and one-shot-event misses others pass over.

Review prompt shape (all reviewers): name the exact files to read (spec + repo contract file like CLAUDE.md), give 4-6 review axes, cap findings (10-15), require `VERDICT: SHIP | REQUEST CHANGES` + findings tagged MAJOR/MINOR/NIT each with a concrete suggested fix, no code proposals.

## Fold findings into the next spec version

- Fix every BLOCKING/MAJOR finding in one pass — don't cherry-pick; batch cheap MINOR/NIT fixes into the same pass or defer them. Add a **revision note** at the top mapping which round's findings drove the change; this makes the next confirm round cheap to verify.
- When a finding is declined, state the decision IN the spec (a decision line where the question arose), not just in chat — reviewers re-raise unstated decisions every round.
- Re-check the spec for contradictions introduced by the fixes BEFORE dispatching the confirm round: fix rounds commonly create new conflicts between sections that no reviewer explicitly flagged.
- Keep the spec as ONE versioned file with revision notes, not v1/v2 file copies — reviewers read the current version; the revision note is the history.

## Confirm rounds: RESOLVED / RESOLVED-BUT / NOT-RESOLVED + NEW hunt

Each confirm-round prompt:
1. Lists every prior finding (number, tag, one-line summary — the reviewer has no memory of prior rounds).
2. Asks per finding: `RESOLVED` | `RESOLVED-BUT <residual>` | `NOT-RESOLVED — <why>`.
3. Demands a NEW-problem hunt: "did the revision introduce new problems (contradictions between sections, untestable criteria, over-engineering)? List as NEW-N with severity, or 'no new problems'." This is mandatory — pure-verify prompts miss fix-induced contradictions, which are common (4 new MAJORs appeared in one round after all 10 prior findings resolved).
4. Ends with VERDICT + one line of justification.

- A RESOLVED-BUT residual that is BLOCKING/MAJOR is fix-in-this-pass; a MINOR one is recorded as a decision line or deferred — never silently accepted (an unstated one resurfaces next round).
- Reproduce reviewer claims against the actual spec text before folding them in — a mis-cited line number once pointed at the wrong finding; the claim was real but lived elsewhere. Triage scope yourself rather than accepting findings as written.
- Findings should narrow each round. If they grow, a fix round is thrashing — stop, re-read the spec yourself, and decide with the user.
- Convergence across 11 user stories took 4 confirm rounds with ONE reviewer (Muse spark-1.3, effort=high, artifact-by-path prompts, bounded tool calls). Muse confirmed with per-finding RESOLVED/NOT-RESOLVED verdicts + NEW-N hunt each round; the NEW-N hunt caught real fix-induced regressions in every round that introduced them (skip-semantics contradiction, kv-read conflict, dedup-wakeup ambiguity). When a NEW finding is genuinely about the fix's own wording rather than the design, fix in place and re-confirm cheaply.

## Ship criteria

Ship after one confirm round reports: no open BLOCKING/MAJOR finding (MINOR/NIT resolved, accepted or deferred), no BLOCKING/MAJOR NEW-N, VERDICT: SHIP. Do not require both original round-1 reviewers to independently SHIP — one reviewer running confirms to convergence is sufficient when round 1 already converged on the same blockers from both reviewers. Re-running the full review for a redundant second SHIP adds cost, not information.

Remaining [NEEDS CLARIFICATION] markers are acceptable at SHIP when the reviewer explicitly says they don't block planning; default each one to a sensible default in the implementation plan and mark it as a default, not a decision.

Budget: the round budget and stop conditions are the shared ones in `confirm-round-protocol.md` (spec reviews use the tier of the work they specify). MINOR/NIT findings and contradictions outside the revised sections go to the deferred list; they do not keep the loop open.

## Artifacts and ledger

Durable record = the spec's revision notes + a per-round ledger in chat (round, reviewer, findings, verdict) — the user pings 'status' and expects current round state instantly. Keep confirm prompts/logs under /tmp per round; do not commit them.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall