| name | implement |
| description | Run implementation phase against an approved plan. Use after /plan has been approved by the human. Triggers on /implement, "implement the plan", "code this up", "execute the plan". |
/implement
Implement the slice against the approved plan. The plan was gated; you follow it.
When to use
- A plan exists (output of
/plan) and the human approved it.
- A worktree + branch is set up for this slice.
- Self-QA tools are ready (
pnpm checks or equivalent).
Steps
- Re-read the spec, the plan, the canonical example, and AGENTS.md / CLAUDE.md.
- Implement strictly to the plan:
- Files in plan's "Files to add/modify" only.
- Mimic the canonical pattern.
- No abstractions beyond what the plan justified.
- No defensive code.
- Mid-implementation, if you discover the plan is wrong:
- Stop.
- Summarize the issue.
- Ask the user. Do not improvise.
- After code is written, run self-QA:
pnpm checks
Fix any failures and re-run until green.
- For UI changes, additionally:
- Open in browser via Playwright skill if available
- Walk happy path
- Take screenshot
- Verify no console errors
- Once green, hand off to
/review.
Hard rules
- Plan is the contract. Don't expand scope, don't add files, don't add abstractions not in the plan.
- No --no-verify on commits.
- Don't commit unless asked. User will instruct.
- Don't claim "done" without
pnpm checks green.
- If you can't browser-test a UI change, say so; don't claim it works.
Next step
/review (anti-overengineering review subagent).