TL;DR

12-priority verification chain plus environment pre-flight. Blocking: security (critical), lint (exit 1), tests (exit 1), and strict deliverable failures. Advisory: tech debt, UX audit, a11y, Lighthouse, Playwright, suggestions, impact prediction, quality trend. Run decision tree to select checks. Fix blockers before completion.

Execution Quality Gate

Activation

1. Activate during final $gate steps.
2. Activate on explicit $codex-execution-quality-gate or $gate.
3. Run before saying work is complete.
4. Activate on $pre-commit or "check before commit".
5. Activate on $smart-test or "which tests to run".
6. Activate on $suggest or "suggest improvements".
7. Activate on $impact or "what will this affect".
8. Activate on $quality-record to record quality trend snapshot.
9. Activate on $quality-report to generate trend report.
10. Activate on $ux-audit.
11. Activate on $a11y-check.
12. Activate on $lighthouse <url>.
13. Activate on $e2e check.
14. Activate on $e2e generate <url>.
15. Activate on $e2e run.
16. Activate on $codex-doctor or $doctor.
17. Activate on $setup-check.
18. Activate on $install-hooks or "install git hooks".
19. Activate on $install-ci or "install CI gate".
20. Activate on $output-guard or $guard or "check if this output is too generic".
21. Activate on $editorial-review or $editorial or "make this read less like AI / more like a human deliverable".
22. Activate on $check or "run quick gate".
23. Activate on $check-full or "run full gate".
24. Activate on $check-deploy or "run deploy gate".

Decision Tree Routing

Task type -> Pre-flight/setup?
    |- Yes -> run: doctor
    `- No -> Code change?
        |- Yes -> What kind?
        |   |- New feature -> run: pre_commit_check + smart_test_selector + predict_impact
        |   |- Bug fix -> run: pre_commit_check + smart_test_selector
        |   |- Refactor -> run: tech_debt_scan + pre_commit_check
        |   `- UI change -> run: ux_audit + accessibility_check + pre_commit_check
        |
        |- Deploy/ship? -> run: security_scan + lighthouse_audit + playwright_runner
        |
        |- Review/audit? -> run: quality_trend + suggest_improvements + tech_debt_scan
        |
        |- Written deliverable quality? -> run: output_guard
        |
        `- No code -> skip quality gate

Runtime Enforcement

• Advisory mode is the default: the AI must decide which gate scripts to run and report fresh evidence before completion.
• auto_gate.py is the single orchestration entry point when the user wants quick, full, or deploy-focused checks without remembering individual script names.
• Runtime enforcement is available through install_hooks.py, which installs a managed pre-commit hook that runs security_scan.py and pre_commit_check.py automatically before commit.
• Use --with-lint-test only when the team wants heavier local enforcement that also runs run_gate.py.
• Use install_ci_gate.py to generate CI enforcement for GitHub Actions or GitLab CI when the project should block pushes and pull requests in automation as well as locally.
• Hooks and CI templates are additive. They do not replace the normal gate decision tree, and uninstall should remove only the managed CodexAI block.
• If runtime enforcement is not installed, the AI must continue to self-enforce the gate manually.

Rules

• Follow the verification order, execution flow, and blocking policy in references/gate-execution-flow.md.
• Blocking failures must be fixed and rerun; warning-only signals may proceed only with explicit user-facing warnings.
• Run --help before invoking a script, treat scripts as black-box helpers, and inspect source only for customization or bug fixing.
• Use with_server.py only when runtime audits need controlled startup and shutdown.
• If user says skip gate or force complete, comply and warn: "Quality gate skipped. Lint/test/security status is unknown."
• For each invoked script, capture output, summarize passes/warnings/blockers, ask whether to fix blocking errors when present, and rerun after fixes to verify.
• Xem skills/.system/REGISTRY.md để biết đường dẫn đầy đủ.

Reference Files

• references/gate-policy.md: blocking vs warning rules for gate decisions.
• references/improvement-suggester-spec.md: post-task suggestion behavior and presentation protocol.
• references/impact-predictor-spec.md: pre-edit blast-radius analysis guidance.
• references/quality-trend-spec.md: periodic quality trend workflow and interpretation.
• references/output-guard-spec.md: heuristics for detecting generic filler and weak evidence in deliverables.
• references/editorial-review-spec.md: rubric for checking whether a deliverable reads like a human, accountable engineering artifact.
• references/ux-audit-spec.md: static UX audit checks and scoring.
• references/accessibility-check-spec.md: WCAG static checks and compliance scoring.
• references/lighthouse-audit-spec.md: Lighthouse wrapper behavior and graceful fallback.
• references/playwright-runner-spec.md: Playwright check/generate/run behavior.
• references/run-gate-spec.md: lint and test orchestration behavior for pass/fail decisions.
• references/security-scan-spec.md: severity-aware security scan integration and blocking criteria.
• references/bundle-check-spec.md: dependency and bundle signal interpretation guidance.
• references/script-commands.md: script CLI examples, including runtime hook and CI installers.
• references/pre-commit-check-spec.md: staged-file fast feedback behavior and blocker handling.
• references/smart-test-selector-spec.md: test selection strategy for changed files.
• references/tech-debt-scan-spec.md: advisory technical debt signal interpretation and prioritization.
• references/gate-execution-flow.md: verification order, execution steps, overrides, and output-handling rules.
• references/output-schemas.md: JSON status contracts for gate helper scripts, including hook and CI installers.