moonlight-lupin/agent-skills / plugins/dev-pipeline/skills/review-commit
review-commit
VerifiedUse 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.
| File | Kind | Size |
|---|---|---|
| SKILL.mdexecutable | skill md | 6.8 KB |
| references/background-build-polling.mdexecutable | script | 3.9 KB |
| references/confirm-round-protocol.mdexecutable | script | 12.3 KB |
| references/dispatch-result-contract.mdexecutable | script | 3.5 KB |
| references/feature-build-brief.mdexecutable | script | 2.9 KB |
| references/fix-brief-template.mdexecutable | script | 10.5 KB |
| references/persistent-loop-plan.mdexecutable | script | 6.9 KB |
| references/release-hygiene.mdexecutable | script | 3.9 KB |
| references/reviewer-dispatch.mdexecutable | script | 6.0 KB |
| references/spec-convergence.mdexecutable | script | 5.2 KB |
| references/two-axis-single-pass.mdexecutable | script | 9.9 KB |
| references/workflow-steps.mdexecutable | script | 18.9 KB |
| scripts/attest_file.pyexecutable | script | 2.7 KB |
| scripts/capture_dispatch_result.pyexecutable | script | 3.6 KB |
| scripts/loop_plan_check.pyexecutable | script | 3.8 KB |
references/reviewer-dispatch.md
View on GitHub# Reviewer Dispatch — running the builder and reviewer CLIs
How to dispatch the lanes in `orchestrator/references/lanes.md` so a round
produces a verdict instead of an environment error. Model names live in
`lanes.md`; this file uses `$BUILDER_MODEL`, `$REVIEWER_A_MODEL` etc. for them.
(Extracted in v0.2 from the old `build-review-loop.md` and
`codex-confirm-loop.md`.)
## Prerequisites
- Cursor CLI (builder): binary is `agent` (`~/.local/bin/agent`), not
`cursor`. Auth: `CURSOR_API_KEY` in `~/.hermes/.env`.
- Codex CLI (reviewer A): `npm install -g @openai/codex`; Codex OAuth or
`OPENAI_API_KEY`; needs `pty=true` and a git repo.
- Claude Code (reviewer B): `claude -p`.
- Muse (reviewer C): `muse exec --trust-workspace`.
- Check before the first dispatch: `which agent codex claude muse`.
- Auth failures that repeat: stop retrying and switch per `cli-auth-fallback`.
## The review input is a pre-generated diff
Don't point reviewers at the tree. Export the exact delta and reference it:
```bash
git diff <base>..<round-commit> > <repo>/review-diff-<proj>.diff # gitignored
codex exec --sandbox read-only -m "$REVIEWER_A_MODEL" -c model_reasoning_effort="high" \
"Read ONLY ./review-diff-<proj>.diff (plus the current files it touches).
<confirm-round prompt per confirm-round-protocol.md>. Review only, change nothing."
```
Bounded context, no re-litigating untouched files, and the diff doubles as
the fix brief's evidence. Muse needs this shape to finish at all. R1 uses the
change's full diff; confirm rounds use the FIX diff
(`confirm-round-protocol.md` § Scope).
Write the review prompt to a file inside the repo cwd (gitignored) and pass
its path — Muse cannot read `/tmp`, and `claude -p` reads only under its cwd.
## Working directory and sandbox
- `codex exec` refuses to run outside a git repo ("Not inside a trusted
directory", exit 1 in < 2 s). Run it from the repo root; reference the diff
by absolute path.
- `claude -p` reads only files under its allowed dirs. Run it from the
directory holding the diff.
- Codex sandbox values are exactly `read-only`, `workspace-write`,
`danger-full-access`. `read-only` cannot run pytest; `workspace-write` can
hang inside slow DB-backed suites. Default: read-only static review with
"do NOT run the test suite" — the orchestrator runs the suite.
- An exit with only an environment error is a failed dispatch, not a clean
review. Never record "no findings" from it.
## Timing
- Codex confirm on a ~2k-line diff: ~10 min. Codex + Muse in parallel at high
effort: ~15-20 min typical; a Muse confirm with a long finding list can run
~50 min. Don't diagnose a hang before ~60 min; Muse finishes last.
- Opus high: 5-8 min per review.
- One timer per background process. A late timer is "stale — superseded";
take elapsed time from `ps -o lstart -p <pid>`, not from counting timers.
- Background polling cadence: `background-build-polling.md`.
## Reading each reviewer's output
- **Reviewer A (Codex):** executes probes from the diff — name the exact
probes to re-run in the confirm prompt.
- **Reviewer B (Opus):** reads deeply; ask for a per-ID fix-map table and a
must-fix vs low-priority split.
- **Reviewer C (Muse):** verdict block of per-item RESOLVED/NOT-RESOLVED with
an ID map, then new issues (severity + file:line), then what it ran. Its
NOT-RESOLVED list can include items it scoped out of its own R1 — check the
ID map before treating them as regressions. New MAJORs are usually real;
MINORs skew to publication/portability.
- Different reviewers find different bug classes — that is why R1 uses
several. It is not a reason to keep every reviewer on every confirm round
(`confirm-round-protocol.md` § Who reviews each confirm round).
## Reviewer death, usage limits, substitution
- Usage limit mid-run ("retry at HH:MM"): don't rewrite the prompt. Harvest
the partial log as leads to reproduce (never as findings), note the reset,
re-dispatch the IDENTICAL prompt after it.
- Opus 529 overload: retry with `--fallback-model sonnet`.
- Builder `resource_exhausted`: drop to the builder fallback in `lanes.md`.
- "Assess yourself instead of <reviewer>": do the confirm yourself against
the prior round's exact finding list (probe each open item, per-item
verdicts, check fix-induced regressions), record who was substituted and on
whose instruction, and cancel any timer waiting on that reviewer.
- A lane is DEAD (tool gone, outage beyond its fallback, no usage window)
during a CONFIRM round: record the outage in the ledger, then substitute
the dead lane's coverage per
`review-lessons/references/generated-frontmatter-and-reviewer-outages.md`
§1 (cross-family substitution — one lane re-probes the dead lane's prior
cases under its own harness). Model names in that lesson are historical;
`orchestrator/references/lanes.md` wins.
- **At the GATE a dead lane cannot be substituted.** The gate requires every
lane's own SHIP on the net diff (`confirm-round-protocol.md` §
Three-reviewer deployment gate). If a lane is dead at gate time: record the
outage, hold the deploy, and re-dispatch that lane when its tool or usage
window returns. Only the USER can approve shipping with a lane's gate SHIP
missing, in writing, recorded in the ledger.
- Killing a respawning `codex exec` worker: loop `os.kill` over `pgrep -f`
output until it returns empty; a single `kill <pid>` loses to the respawn.
## Bot Mode (optional)
Two bot profiles can run the lanes with persistent context:
| Bot | Profile | Role |
|---|---|---|
| Coder | `coder` | builder lane — drives Cursor CLI with `$BUILDER_MODEL` |
| Code Reviewer | `code_reviewer` | reviewer lanes A/B/C — Codex, Claude Code (Opus), Muse |
```bash
hermes -p coder chat --in ~ -c "Bot Chat" --create-if-missing -Q -q \
"Message from 🤖 hermes (@hermes): <task>"
```
Run with `background=true, notify_on_complete=true`. Use `delegate_task` for
quick same-model subtasks; bots for cross-model build/review with memory
across rounds. The confirm-round protocol applies unchanged.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| 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: 198k | Updated | Claude CodeCodex | Verified | Install |
| 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: 119k | Updated | Claude CodeCodex | Verified | Install |
| 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: 118k | Updated | Claude CodeCodex | Verified | Install |