AI Release

Release Gate

Aggregated GO/NO-GO release readiness gate. Checks every quality dimension against manifest.yml thresholds and produces a verdict with evidence. Run before version tagging or merge-to-main.

When to Use

• Pre-release milestone, version tagging, merge-to-main decision.
• NOT for code quality only -- use /ai-quality.
• NOT for security only -- use /ai-security.

Process

Phase 1: Gate Dimensions

1. Coverage -- verify test coverage meets threshold.

   • Run pytest tests/ --cov --cov-report=term-missing.
   • Gate: coverage >= threshold from manifest.yml (default 80%).

2. Security -- verify zero medium+ findings.

   • Run gitleaks protect --staged, semgrep scan --config auto ., pip-audit.
   • Gate: zero critical/high findings. Medium findings documented.

3. Tests -- verify all tests pass.

   • Run pytest tests/ -v.
   • Gate: 100% pass rate. Zero skipped governance tests.

4. Lint -- verify clean lint and format.

   • Run ruff check ., ruff format --check ..
   • Gate: zero unfixable lint errors.

5. Dependency vulnerabilities -- verify clean dependency tree.

   • Run pip-audit --strict.
   • Gate: zero known vulnerabilities. Accepted risks must be in decision-store.json.

6. Type checking -- verify type correctness.

   • Run ty check src/.
   • Gate: zero errors.

7. Documentation coherence -- verify docs are current.

   • CHANGELOG.md has [Unreleased] entries or versioned section.
   • README.md reflects current features.

8. Packaging integrity -- verify distribution builds.

   • Run uv build or equivalent.
   • Gate: wheel builds cleanly, no missing files.

Phase 2: Aggregate Verdict

9. Determine verdict:

   • GO: all dimensions pass.
   • CONDITIONAL GO: all pass except non-critical items with risk acceptance in decision-store.json.
   • NO-GO: one or more blocking issues without risk acceptance.

10. Produce closure path (for NO-GO):

    • List specific blockers with fix suggestions.
    • Prioritize by blocking severity.
    • Estimate effort per fix.

Output Contract

## Release Readiness: [version]

### Verdict: GO | CONDITIONAL GO | NO-GO

| Dimension | Status | Detail |
|-----------|--------|--------|
| Coverage | PASS 87% (>= 80%) | -- |
| Security | PASS | 0 findings |
| Tests | PASS | 142/142 pass |
| Lint | PASS | 0 errors |
| Dependencies | PASS | 0 vulns |
| Types | PASS | 0 errors |
| Docs | PASS | CHANGELOG current |
| Packaging | PASS | wheel builds |

### Blockers (if NO-GO)
- [Issue, severity, fix suggestion]

### Residual Risk
- [Risk acceptances from decision-store.json]

Quick Reference

/ai-release              # full GO/NO-GO gate
/ai-release v2.0.0       # gate for specific version
/ai-release --check-only # report without blocking

Common Mistakes

• Running release gate on a dirty working tree -- commit first.
• Ignoring CONDITIONAL GO risks -- each must have a decision-store.json entry.
• Skipping packaging integrity -- wheel build failures are release blockers.

Integration

• Aggregates results from /ai-security, /ai-quality, /ai-test.
• Reads thresholds from manifest.yml.
• Risk acceptances from state/decision-store.json.

References

• .claude/skills/ai-security/SKILL.md -- security gate.
• .claude/skills/ai-governance/SKILL.md -- compliance and risk.
• .ai-engineering/manifest.yml -- quality thresholds. $ARGUMENTS