sk-review-plan

Read-only review of a technical plan. Use when sk-planning has drafted a plan, or the user passes a plan and asks for a review.

SKILL.md

GitHub

Plan Review

Evaluate a technical plan before implementation. Flag correctness blockers, gaps, and unnecessary complexity.

When to use

  • After sk-planning writes a draft.
  • When the user pastes a plan and asks for a review.
  • Before presenting a masterplan or sub-plan to a stakeholder.

Input

Paste the full plan text into the prompt. Do not point the subagent at plan files. Include repo constraints, user scope, masterplan ↔ sub-plan linkage (if split), and known validation gaps.

Output contract

Return ONLY the findings block. No prose intro.

Output ONLY in this format (no prose intro, no sections swapped).

### Plan Review Findings

section:line: <emoji> <tag>: <problem/over-engineering>. <fix/replacement>.
totals: N🔴 N🟡 N✂️ N🪵 N⚡ N🔵 N❓ | net: -<N> lines/steps possible

Or exactly when no findings:

### Plan Review Findings

Lean & valid. Ship.

Rules for Plan Review Findings:
- Sort findings section → line ascending
- One finding per line; problem then fix separated by ". "
- Location: section = exact ## heading name; line = line number within that section (1-based from heading line), or — when issue spans whole section
- Action tags (literal after colon):
  - 🔴 bug: correctness blocker, contradiction, broken logic, missing critical step
  - 🟡 gap: weak testability, missing file/dependency, unclear sequencing, unmitigated risk
  - ✂️ cut: redundant prose, duplicate information, speculative task not required for goal (replacement: nothing)
  - 🪵 yagni: over-engineered scaffolding, premature generalization, unnecessary abstraction (replacement: inline or drop)
  - ⚡ simplify: over-complicated sequencing, multi-tier nesting, complex pattern (replacement: simpler layout or step)
  - 🔵 nit: wording, formatting, minor presentation tweak
  - ❓ question: unclear intent, missing context, needs author clarification
- Focus: self-containment, executability, TODO atomicity/testability, scope/acceptance alignment, missing files/dependencies, contradictions, realistic sequencing, untestable verification, risks without mitigation, Complexity Check accuracy, masterplan ↔ sub-plan consistency, and ponytail complexity reduction. Do NOT perform pure micro prose edits unless they reduce scope, file count, or steps.

End with exactly one line:
- totals: N🔴 N🟡 N✂️ N🪵 N⚡ N🔵 N❓ | net: -<N> lines/steps possible
- Or: Lean & valid. Ship.

Tags

  • 🔴 bug: correctness blocker, contradiction, broken logic, missing critical step
  • 🟡 gap: weak testability, missing file/dependency, unclear sequencing, unmitigated risk
  • ✂️ cut: redundant prose, duplicate information, speculative task not required for goal
  • 🪵 yagni: over-engineered scaffolding, premature generalization, unnecessary abstraction
  • ⚡ simplify: over-complicated sequencing, multi-tier nesting, complex pattern
  • 🔵 nit: wording, formatting, minor presentation tweak
  • ❓ question: unclear intent, missing context, needs author clarification

Review criteria

  1. Self-containment: executable using only the plan plus a normal repo checkout.
  2. Executability: every step can be done without guessing missing context.
  3. TODO atomicity/testability: each TODO is one focused, testable action.
  4. Scope/acceptance alignment: the plan matches the goal and acceptance.
  5. Missing files or dependencies: the Files section covers every path that needs to change.
  6. Contradictions: sections do not contradict each other.
  7. Realistic sequencing: dependencies and ordering are possible.
  8. Verifiable: verification steps are concrete and runnable.
  9. Risk coverage: risks are acknowledged or mitigated.
  10. Complexity Check accuracy: counts and Proceed/Split decision are honest.
  11. Masterplan ↔ sub-plan consistency: sub-plans repeat every fact they need from the masterplan.
  12. Ponytail complexity reduction: the plan is as small as it can be while still correct.

Triage

Label each finding valid, false_positive, or unvalidated with a one-line reason. If zero valid findings remain, output exactly Lean & valid. Ship..

Rules

  • Read-only: do not edit the plan or any repo files.
  • Single block: emit exactly one ### Plan Review Findings block.
  • Sort findings by section, then line.
  • One finding per line; problem then fix, separated by ". ".
  • Do not add new sections or rewrite the plan.
  • Minor wording issues are not findings unless they mislead.