implement-planlisted
Install: claude install-skill nuttaruj/rolepod
# Implement Plan
Execute the approved plan with discipline: TDD where it matters, surgical edits, fresh-context review of delegated work, worktrees only when filesystem isolation is real.
## Iron Rule
<EXTREMELY-IMPORTANT>
1. NEVER edit code without reading the plan and the touched files first.
2. NEVER expand scope mid-implementation. New idea → one line under the plan's `## Follow-ups`, then finish the planned task.
3. ALWAYS write the failing test first for bug fixes and high-risk-surface work.
4. NEVER delegate without a written task scope and a clear done criterion.
5. CONTINUOUS execution between tasks AND between plan phases — no "should I continue?" check-ins, no progress summaries, never end the turn mid-plan: an ended turn is a stop however it is worded. Stop only on a BLOCKED, spec / plan gap, or scope ambiguity that SURVIVES a re-read of the plan and the touched files (BLOCKED: plus a variable change) — a wrinkle you can settle yourself is never grounds to stop.
6. Forced to end anyway (usage limit / context / user stop) → the last act is one line under the plan's `## Changes during build`: stopped after Task N · next Task M · how to start the env — the next session reads it before anything else.
</EXTREMELY-IMPORTANT>
## Skip when
- A question only · the plan is still vague → `write-plan` first.
## Boundary
Owns: executing the approved plan — reading touched files, surgical edits, task-level tests, bounded delegation.
Does not own: changing product scope