| name | spec-verify |
| description | Run every gate and check a spec's acceptance criteria one by one. Use when asked to "verify a spec", "check if this is done", "run the gates", "is spec NNN complete", or before marking any work finished. Produces a criterion-by-criterion table with evidence. |
Verify a spec
Verify specs/NNN-name/spec.md is genuinely satisfied.
Portable equivalent of the /spec-verify prompt file. Prompt files work
only in the VS Code Chat view; this skill works there and in the Copilot CLI
and the coding agent.
1. Run every gate, show every output
cd apps/api
uv run ruff check .
uv run ruff format --check .
uv run mypy src
uv run pytest -q
uv run pytest -m golden -v
cd apps/web
pnpm exec tsc --noEmit
pnpm exec vitest run
pnpm exec vite build
Paste the real output. Not a summary of it. If something fails, that is the
finding — report it and stop pretending otherwise.
2. Walk the acceptance criteria individually
For each numbered criterion in spec.md, state one of:
- ✅ Met — and name the specific gate, test, or command that proves it
- ❌ Not met — and say what is missing
- ⚠️ Unverifiable — and say why the criterion cannot be checked as written
A criterion is not met because the code looks like it should work. It is met
because something ran and passed.
3. Check the repository rules
- No credentials or PII introduced anywhere in the diff
analytics/ still free of AI and network callsagent/ still reaching data only through analytics/- Money still integer cents; Postgres sums still coerced with
int()
- Ranking still on
peer_index
- Market payloads still declaring
is_synthetic
gitleaks detect --no-git --redact
4. Report
Give a table: criterion → status → evidence.
Then one plain sentence: is this spec done, or not? If any criterion is
unmet, the answer is "not", regardless of how much of it works.
Do not mark anything complete in spec.md that you have not proven here.
