spec-quality-gate

Run the pre-lint script against the spec file first to catch mechanical format failures, then dispatch the spec-quality-reviewer agent with the spec path. Re-run both after each round of author fixes until the reviewer returns PASS.

The bash script checks YAML frontmatter fields (spec_id, title, status, north_star), sequential REQ numbering without gaps or duplicates, REQ block completeness, Test Coverage Matrix consistency, Out of Scope and Non-Functional Requirements sections, and residual placeholders like TBD.

It runs after brainstorming writes the spec and before high-level-design or writing-plans starts. No planning may proceed without a passing gate, and a second traceability gate runs after implementation before merge.

The loop has no fixed cycle cap. Escalation to a human occurs only on genuine non-convergence, meaning a re-run returns the same failure the author already attempted to fix with no forward progress, indicating a design problem the gate cannot resolve.

No. The script only covers deterministic Section 1 format checks and the placeholder scan. Judgment checks such as requirement clarity and north-star alignment still require the spec-quality-reviewer agent.