implement

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

Before implementing, verify the work is written down. This is a hard precondition: run the flow-gates precondition check first and stop unless it passes.

```bash
node scripts/flow-gates/precondition.mjs
```

If it exits non-zero, there is no local spec/Issue file — stop and ask the user to run /to-spec (then /to-tickets) before implementing. If the check is not installed, stop and ask the user to run /setup-matt-pocock-skills to install the flow-gates — do not fall back to a manual check. A spec that exists only on a remote issue tracker is not a local file the check can see; the file-based gate requires a local handoff artifact. Implementing without a written-down spec/Issue is out of scope.

Implement the work described by the user in the spec or Issues.

Use /tdd where possible, at pre-agreed seams.

Run the repository-declared typecheck command regularly when one exists. If the repository has none configured, record "not configured" and do not introduce a checker, configuration, or dependency unless the ticket authorizes it. Run single test files regularly, and the full test suite once at the end.

When marking a ticket's acceptance criteria done, record the actual command run, its exit code, test counts, and any warnings. Never mark a criterion done without running its verification — and not if the run produced warnings or could not be reproduced. New warnings, or warnings in ticket-owned tests, block completion; pre-existing unrelated warnings are recorded but not fixed unless the ticket scopes them.

Stay inside the ticket's declared scope: nothing from its Out of scope (deferred) section, no new dependencies, checkers, configs, or repo-level workflows unless the ticket explicitly authorizes them. Respect the ticket's global invariants even where another ticket owns the full contract.

Capture the base commit before editing. Implement the ticket, run its acceptance checks, and record completion evidence. Commit only ticket-owned changes.

Run /code-review against base...HEAD. Fix every blocking finding, re-run the affected checks, commit the fixes, and repeat the review against the same base until no blocking findings remain. After the last fix, re-run all ticket acceptance checks and the full repository suite.

Then mark the ticket completed (flip its Implementation status), commit that status/evidence update, and run one final review against the same base.

If the final review finds any blocking issue, immediately flip Implementation status back to pending, fix the issue, re-run all ticket acceptance checks and the full repository suite, commit, and repeat the final review against the same base. A ticket is completed only while the latest HEAD has a clean final review.

From showgp/skills-with-gate, MIT.

More from showgp/skills-with-gate