spec-kit-checklist

Audit requirement quality across change artifacts. This is a requirements linter, not a test runner.

Input: Optionally specify a change name. If omitted, infer from context or ask.

Steps

1. Load artifacts

   Load what exists under specs/changes/<name>/:

   • spec.md, plan.md, tasks.md

   Also load FRAMEWORK.md for governing principles alignment.

2. Ask up to 3 clarifying questions (skip if context is clear)

   Focus on: scope, risk prioritisation, audience, boundary exclusions.

3. Generate requirement quality checklists

   Create or update files in specs/changes/<name>/checklists/ grouped by domain (e.g., api.md, ux.md, security.md).

   Each checklist item must:

   • Use question format: "Are [requirement aspect] defined for [scenario]?"
   • Include a quality dimension tag: [Completeness], [Clarity], [Consistency], [Measurability], [Gap], [Ambiguity], [Conflict]
   • Reference the source: [Spec §X] or [Plan §X]
   • Never duplicate existing checklist IDs — continue numbering from the last CHK-N

   Quality dimensions to cover:

   • Completeness — are all scenarios and edge cases specified?
   • Clarity — are terms unambiguous and objectively measurable?
   • Consistency — do requirements contradict each other?
   • Measurability — can success criteria be verified without interpretation?
   • Coverage — are all user stories traceable to tasks?
   • Non-functional — are performance, security, and scale requirements present?

   ❌ Prohibited: "Verify button clicks", "Test API returns 200" — these are implementation tests. ✅ Required: "Is fallback behavior defined when X fails?", "Can 'fast' be objectively measured?"

4. Report

   Produce a summary:

   • Total items by dimension
   • High-priority gaps
   • Items that block proceeding to the spec-kit-plan or spec-kit-implement skills

Guardrails

• Never delete existing checklist items — only append.
• Maintain ≥80% traceability across checklist items.
• Flag FRAMEWORK.md principle gaps as CRITICAL.
• This skill audits requirements, not code behavior.