AI Governance

Governance

Compliance validation for regulated industries. Modes: compliance (quality gates), ownership (boundary verification), risk (decision-store lifecycle), integrity (framework consistency). Default: compliance.

When to Use

• Governance audit, pre-release check, post-install verification.
• NOT for code quality -- use /ai-quality.
• NOT for security scanning -- use /ai-security.

Modes

compliance -- Quality Gate Validation

Validate that rules in framework-contract.md are enforced.

1. Hook enforcement -- verify required hooks exist in .git/hooks/, are executable, contain no --no-verify escapes.
2. Check coverage -- for each stack in enforcement.checks, confirm tool is configured and callable.
3. Non-negotiables -- walk standards.non_negotiables, trace enforcement chain: manifest -> hook -> CLAUDE.md.
4. CI workflows -- verify enforcement.ci.required_workflows exist under .github/workflows/.
5. Security contract -- gitleaks in pre-commit, semgrep in pre-push, dependency audit per stack.

ownership -- Boundary Validation

Verify files live in correct ownership zones.

1. Zone mapping -- load ownership.model from manifest. Build zones: framework-managed, team-managed, project-managed, system-managed.
2. File placement -- scan .ai-engineering/, verify each file maps to exactly one zone.
3. Modification history -- git log framework-managed files, confirm only framework commits touched them.
4. Update rule compliance -- team/project paths never overwritten by automation.

risk -- Risk Acceptance Lifecycle

Sub-modes: accept, resolve, renew.

Accept: record time-limited risk in decision-store.json.

• Classify finding, determine severity, register with mandatory follow_up_action.
• Auto-expiry: Critical 15d, High 30d, Medium 60d, Low 90d.

Resolve: close after remediation.

• Validate fix committed, scan clean, no regression.
• Mark remediated (preserved for audit trail, not deleted).

Renew: extend before expiry (max 2 renewals).

• Check eligibility (renewal_count < 2). Require justification.
• Create new decision with renewed_from reference.

integrity -- Framework Consistency

Validate manifest claims match disk reality.

1. Manifest counters -- compare governance_surface.agents.total and skills.total against actual file counts.
2. Agent-skill references -- verify every path in agent references.skills resolves to an existing SKILL.md.
3. State file schemas -- confirm state/ files are valid JSON/NDJSON with required keys.
4. Command file existence -- verify each SKILL.md has valid YAML frontmatter.

--report -- Formal Report

Generate structured compliance report suitable for audit:

# Governance Report: [mode]

## Score: N/100
## Verdict: PASS (>=90) | WARN (>=70) | FAIL (<70)

## Findings
| # | Severity | Category | Description | Location | Remediation |

## Gate Check
- Blocker: N (threshold: 0)
- Critical: N (threshold: 0)

Scoring: start at 100. Deduct: blocker -25, critical -15, major -5, minor -1. Floor at 0.

Quick Reference

/ai-governance              # compliance mode (default)
/ai-governance all          # all modes
/ai-governance risk accept  # accept a new risk
/ai-governance risk resolve # close a remediated risk
/ai-governance integrity    # framework consistency check
/ai-governance --report     # generate formal report

Common Mistakes

• Running governance mid-implementation -- best between phases or before releases.
• Accepting risk without follow_up_action -- mandatory field.
• Exceeding 2 renewals -- remediation becomes mandatory.

Integration

• CLI layer: ai-eng validate --category <mode>, ai-eng doctor, ai-eng maintenance risk-status.
• Risk acceptances block pre-push when expired.
• Release gate (/ai-release) checks governance status.

References

• .ai-engineering/manifest.yml -- governance non-negotiables and quality thresholds.
• state/decision-store.json -- risk acceptance records. $ARGUMENTS