Quality Gate

Overview

Orchestrate every quality dimension into a single go/no-go verdict. This is the "make the project perfect" pass: it chains the specialist skills (code review, security, performance, metrics) and reduces their output to a decision with blocking items.

Gate dimensions

1. Correctness - code review of the diff or release scope (code-reviewer); tests pass; no known regressions.
2. Security - audit pass (security-auditor) + flaw hunt (vulnerability-scanner); zero unresolved criticals.
3. Performance - no measured regression on hot paths (performance-auditor); budgets respected (bundle size, response times).
4. Reliability - error handling at boundaries, graceful degradation, rollback path exists.
5. Maintainability - lint/type checks clean, no new TODO-debt spike, complexity hotspots justified.
6. Documentation - README/CHANGELOG updated, breaking changes flagged, config documented.
7. Observability - logs at the right level, metrics/health endpoints intact.

Workflow

1. Define the scope: a release tag, a branch diff, or the whole project.
2. Save a checkpoint (/checkpoint save pre-gate) so the state is recoverable.
3. Run each dimension. Delegate to subagents in parallel when available - one per dimension - and consolidate their reports.
4. Score each dimension: PASS / WARN / BLOCK, with evidence.
5. Verdict:
   • Any BLOCK → NO-GO, list blocking items in fix order.
   • Only WARNs → GO with reservations, list follow-ups.
   • All PASS → GO.
6. Output a single gate report: verdict on top, dimension table, then details. Offer to fix blocking items one by one, re-running only the affected dimension after each fix.

Anti-patterns

• Giving GO with an unresolved critical security finding
• Re-running the entire gate after every one-line fix
• Letting one dimension's verbosity bury the verdict
• Skipping the checkpoint before fix sessions