keel-phase-review
Skill9 starsChanged 52 days agoMITView on GitHub
End-of-phase gate — verify working product, tests, docs, ADRs, and handover before committing a phase.
- Reads credentials
In this repository17 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- keel-adoptSkill
- keel-agent-team-createSkill
- keel-agent-team-startSkill
- keel-auditSkill
- keel-autopilotSkill
- keel-compactSkill
- keel-continueSkill
- keel-distillSkill
- keel-handoverSkill
- keel-pilotSkill
- keel-planSkill
- keel-researchSkill
- keel-statsSkill
- keel-teamSkill
- keel-tidySkill
- keel-updateSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Run at the end of a phase (rules.md §2).
≈ “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
/keel-phase-review — end-of-phase checkpoint
Run at the end of a phase (rules.md §2). Report the checklist with pass/fail; fix gaps before proposing a commit.
Checklist:
- Working product — the phase's feature runs; include a one-line "how to test this" for the user. What only a human can verify (browser/UX, prod host, real creds) is handed off as a recipe and tracked in HANDOVER (d) — never silently assumed passed.
- Tests — relevant unit/integration (e2e if needed) written and run via
make test(auto-log →reports/tests/<date>/); the phase's committed summary written toreports/tests/<date>-phase<N>.md(suites, counts, what was verified, and measured durations: suite runtime + the gate's key step timings — WHAT to time is project-specific (E2E latency, first token, query ms, batch throughput, …) — side-by-side with the previous phase's numbers so slowdowns surface at the gate, not in production); new test/fixture files one-lined in their folder README; repeatedly-run corpora (golden / regression / pilot question sets) exist as DATA files with stable ids —tests/fixtures/or a corpus dir indexed fromtests/README.md, never only inline in test code (§10.39 feeds them: every point-fixed failing case joins one); one-line summary in HANDOVER.md (rules.md §2.8); bulk outputs passed the/keel-pilotgates where applicable. Ad-hoc reports (speed tests, golden proofs) complementreports/tests/<date>-phase<N>.md— never replace it. - Architecture — every structural change recorded in
docs/architecture.md: new/changed module rows AND the component overview resynced with them (the module table is the source of truth — a stale top diagram fails this item, same contract as PLAN.md's table → diagram; rules.md §1.6). - PLAN.md (do not skip — this is what marks the phase
done) — the finished phase's gate flipped todone(+ next phase setwip, Current focus updated, diagram regenerated from the table — see/keel-plan). The Stop hookplan-phase-nudgereminds you if you end a turn having cleared awipphase's## Nowwithout flipping it. - ADRs — any significant decision captured as an ADR in
docs/adr/(+ index row). - Docs — CLAUDE.md / docs/user_manual.md updated, or the review states why nothing user-facing
changed. A still-template
user_manual.mdwhile shipped phases have user-facing behavior = FAIL (the SessionStart hook warns on this too — silence is not a pass). - HANDOVER.md — session block added (run
/keel-handover); caps respected (else/keel-distill). - TASKS.md — done items deleted (one-liners in HANDOVER (a));
## Discoveredtriaged. - LESSONS.md — this phase's agreed rules/tests/gotchas written (rules §9.31).
- Audit —
/keel-auditrun over this phase's range via the skill — it writes.claude/last-audit; an inline audit that skips the marker leaves the audit-due nudge firing forever (or consciously skipped with a reason); its approved fixes applied (rules.md §4.11). - Code review —
/code-review(or the project's review flow) run over this phase's diff; each finding triaged through theverifiersubagent (adversarial "try to refute") before any fix is applied: CONFIRMED → fix now · REFUTED → record with its counter-evidence · UNCERTAIN → ask the user (default: don't fix). Reviewer output is a claim, not a verdict (rules.md §4.11); deferred items get one line each inTASKS.md ## Discovered. The phase gate is the review boundary — cheaper than reviewing every session, earlier than reviewing at release. - Secrets —
git diff --cachedreviewed; no.env/secrets staged (rules.md §6.18). - Deps — if dependencies changed: lock refreshed +
pip-auditclean (rules.md §7.23).
Then propose an approved commit + push (rules.md §6.15). Never push without user approval.
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.

