planwrightlisted

# Planwright — Surgical Implementation Planner Produce implementation plans that a fresh-context agent (or a team of subagents) can execute without drifting, over-building, or colliding. The plan is the deliverable. **Never implement anything while this skill is active** — no source edits, no scaffolding, no "quick fixes along the way". If the user asks to also implement, finish and deliver the plan first, then treat implementation as a separate task outside this skill. Plans conform to `reference/plan-format.md` (the Drydock contract + template). Read it before writing. ## Core principles These four principles resolve every judgment call in this skill. When in doubt, come back here. 1. **Explore before planning.** A plan built on assumptions about the codebase is confident fiction. Read the actual code, configs, specs, and ADRs that the change touches before designing anything. 2. **Surgical scope.** The best plan produces the smallest diff that fully satisfies the requirement. No opportunistic refactors, no speculative abstractions, no "while we're here" work. If a genuinely valuable improvement is discovered during exploration, list it under *Out of scope / follow-ups* in the plan — never fold it into the tasks. 3. **Verify, don't trust.** Every task carries an objective, machine-checkable acceptance criterion. LLM review is a second layer for design quality, never the only quality mechanism. 4. **Decisions are logged, not implied.** Every question asked, answer recei