review-gates
Skill0 starsChanged 2 days agoMITView on GitHub
Default review gates for any of Sarah's projects — which review runs when, on which model, at what context cost. Use before every commit, when integrating a helper's lane, and when a batch of PLAN.md items is done. If the project has its own review-workflow skill, follow that too; it wins on conflict.
In this repository10 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- commitSkill
- delegateSkill
- design-reviewSkill
- endSkill
- retroSkill
- reviewSkill
- startSkill
- web-kitSkill
- working-modelSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 0 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Before the commit: the full gate (npm run gate).
- Very specific: line 21
No raw logs, no screenshots.
- Very specific: line 25
Read only git diff <main>...lane/<name> (or the staged diff) plus the helper's report.
- Very specific: line 39
Update PLAN.md and Lessons.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Review gates
Goal: quality at the lowest token cost. Reviews read diffs, not the codebase, and heavy output stays out of the orchestrator's context.
Gate 0 — self-check before every commit (whoever commits)
- While iterating: the project's fast gate (
npm run gate:fastor equivalent). - Before the commit: the full gate (
npm run gate). Only known failures allowed (listed inPLAN.md). - The project's constraints from its
CLAUDE.md. - No raw logs, no screenshots. Helpers report in ≤ 15 lines (see
delegate).
Gate 1 — diff review, per lane or per own commit (orchestrator)
- Read only
git diff <main>...lane/<name>(or the staged diff) plus the helper's report. - Lens: does each commit do what its item says; stays in owned files; project constraints hold; each new test would fail before the change; no duplicate tests.
- Verdict
SHIP/BLOCKED+ ≤ 5 one-line findings. - Blockers go back to the same helper via
SendMessage. Should-fix → follow-up commit or a line inPLAN.mdNext. - Risky task (JS behaviour, accessibility, security): one Sonnet reviewer helper, diff only, blockers only, ≤ 10 lines.
Gate 2 — whole batch, once (Sonnet subagent)
- One Sonnet helper runs the
reviewskill over the batch diff. Blockers only, ≤ 10 lines. - Fix blockers only. No iterate-to-a-score loop; no expert panel unless Sarah asks.
- Visual sign-off is Sarah's: point at the pages to open.
- Update
PLAN.mdand Lessons.
Models
- Helpers and reviewers: Sonnet, passed explicitly. Orchestrator: briefs, Gate 1, merges, docs — judgment only. Haiku only for mechanical edits with no judgment, never for review.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

