← ClaudeAtlas

review-planlisted

Attack a finished plan before anyone files issues from it: whether it covers the design it claims to, whether its acceptance criteria would catch a wrong implementation, whether the scope drifted, and whether the order holds. Use it when someone hands you a plan and wants to know if it is any good — "is this broken down right?", "does this plan actually cover the design?", "these acceptance criteria feel weak", "did we miss anything?", "is this too many issues?" — or when you are about to turn a plan into issues and nobody has challenged it yet. The question it asks is not "does the plan cover the design?" but "would this plan catch the wrong implementation?", and criteria that pass either way are the failure it exists to find. `/plan` already runs it on plans it writes, so the case for reaching for it deliberately is a plan you did not just author. Do NOT use it to write or decompose a plan (`/plan`, `/scope`) or to review code.
tsukumogami/shirabe · ★ 0 · Code & Development · score 67
Install: claude install-skill tsukumogami/shirabe
!`bash ${CLAUDE_PLUGIN_ROOT}/scripts/skill-preflight.sh review-plan 2>&1 || true` # Review Plan Skill `/review-plan` adversarially challenges a complete plan artifact before any issues are created. It runs four review categories against the plan's wip/ artifacts and the upstream design doc, then writes a structured verdict to one of two files depending on outcome. **Writing style:** Read `skills/writing-style/SKILL.md` for guidance. ## Overview The skill sits between `/plan` Phase 5 (Dependencies) and Phase 7 (Creation). It asks not "does the plan cover the design?" but "would this plan catch the wrong implementation?" Four categories map to the three failure modes identified in issue #19: | Category | Name | Failure mode covered | |----------|------|---------------------| | A | Scope Gate | Plan too large or too small for design complexity | | B | Design Fidelity | Plan inherits a contradiction from the design doc | | C | AC Discriminability | ACs pass for the wrong implementation | | D | Sequencing/Priority Integrity | Must-run QA scenarios are deprioritized | ## Execution Modes ### Fast-path (default) Called as a sub-operation by `/plan` Phase 6. One agent evaluates each category. Optimized for latency — same coverage as adversarial mode, lower depth. Invoked via Agent task with: ``` skill: review-plan args: plan_topic: <topic> round: <N> mode: fast-path ``` Each of the four review categories (phases 1–4) runs with a single agent. The agent applies heurist