agentic-governance（Agent 治理）

governance (治理)

Unified skill for constraint governance state, periodic reviews, index generation, round-trip verification, and schema migration. Consolidates 6 granular skills.

Trigger: 定期保守 (periodic maintenance) or HEARTBEAT

Source skills: constraint-reviewer, index-generator, round-trip-tester, governance-state, slug-taxonomy, adoption-monitor (from safety)

Installation

openclaw install leegitw/governance

Dependencies:

• leegitw/constraint-engine (for constraint data)
• leegitw/failure-memory (for observation data)

# Install full governance stack
openclaw install leegitw/context-verifier
openclaw install leegitw/failure-memory
openclaw install leegitw/constraint-engine
openclaw install leegitw/governance

Standalone usage: Index generation and round-trip verification work independently. Full governance features require constraint-engine and failure-memory integration.

Data handling: This skill operates within your agent's trust boundary. When triggered, it uses your agent's configured model for governance analysis and review. No external APIs or third-party services are called. Results are written to output/governance/ in your workspace.

What This Solves

Constraints that never get reviewed become stale. Rules that never get challenged become dogma. This skill manages the lifecycle:

1. State tracking — know which constraints are active, suspended, or retired
2. Periodic reviews — 90-day gates to re-evaluate constraints against current evidence
3. Index generation — dashboards showing constraint health at a glance

The insight: Good governance is proactive. Constraints need maintenance, not just creation.

Usage

/gov <sub-command> [arguments]

Sub-Commands

Arguments

/gov state

/gov review

/gov index

/gov verify

/gov migrate

Configuration

Configuration is loaded from (in order of precedence):

1. .openclaw/governance.yaml (OpenClaw standard)
2. .claude/governance.yaml (Claude Code compatibility)
3. Defaults (built-in)

Core Logic

Governance State Model

┌─────────────────────────────────────────┐
│           GOVERNANCE STATE               │
├─────────────────────────────────────────┤
│ Constraints:                             │
│   - Active: 5                           │
│   - Draft: 2                            │
│   - Retiring: 1                         │
│   - Retired: 12                         │
├─────────────────────────────────────────┤
│ Reviews:                                 │
│   - Due: 2 (approaching 90-day mark)    │
│   - Overdue: 0                          │
├─────────────────────────────────────────┤
│ Health:                                  │
│   - Circuit: CLOSED                     │
│   - Violations (30d): 3                 │
│   - Adoption rate: 85%                  │
├─────────────────────────────────────────┤
│ Alerts:                                  │
│   - [WARN] CON-001 due for review       │
│   - [INFO] 2 new observations eligible  │
└─────────────────────────────────────────┘

Review Cycle

Constraints require periodic review. The review cadence is configurable (default: 90 days):

# .openclaw/governance.yaml
governance:
  review_cadence_days: 90    # Default
  warning_threshold: 15      # Days before due to warn

⚠️ Advisory Only: This review cycle is *not enforced programmatically*. Compliance relies on HEARTBEAT P3 checks and manual diligence. Automated enforcement (/gov review --automated) is planned for future release. See HEARTBEAT.md for current verification schedule.

Adoption Monitoring

Track constraint adoption across sessions:

Slug Taxonomy

Standard slug prefixes for observations and constraints:

Output

/gov state output

[GOVERNANCE STATE]
Updated: 2026-02-15 10:30:00

=== Constraints ===
Active: 5 | Draft: 2 | Retiring: 1 | Retired: 12

=== Circuit Breaker ===
Status: CLOSED (healthy)
Violations (30d): 3

=== Reviews ===
Due: 2 constraints approaching 90-day mark
  - CON-20251120-001: "Always run tests" (day 87)
  - CON-20251125-003: "Lint before commit" (day 82)

=== Adoption ===
Rate: 85% (target: >80%)
Sessions tracked: 47

=== Alerts ===
[WARN] CON-20251120-001 due for review in 3 days
[INFO] 2 observations eligible for constraint generation

/gov review output

[CONSTRAINT REVIEW QUEUE]

Due for review (2):

1. CON-20251120-001: "Always run tests before commit"
   Age: 87 days | Status: active
   Violations (90d): 2 | Overrides: 0
   Adoption: 92%

   Options:
   a) Renew for 90 days: /ce lifecycle CON-20251120-001 active
   b) Begin retirement: /ce lifecycle CON-20251120-001 retiring
   c) Immediate retire: /ce lifecycle CON-20251120-001 retired

2. CON-20251125-003: "Always lint before commit"
   Age: 82 days | Status: active
   Violations (90d): 5 | Overrides: 1
   Adoption: 78%

   [WARN] Below adoption target (80%)
   Consider: Clarify constraint or improve tooling

/gov index output

[INDEX GENERATED]
Path: agentic/INDEX.md
Skills: 7
Updated: 2026-02-15 10:30:00

Contents:
- failure-memory (fm) - Core
- constraint-engine (ce) - Core
- context-verifier (cv) - Foundation
- review-orchestrator (ro) - Review
- governance (gov) - Governance
- safety-checks (sc) - Safety
- workflow-tools (wt) - Extensions

/gov verify output

[ROUND-TRIP VERIFICATION]
Source: docs/constraints/
Compiled: output/constraints/

Status: ✓ IN SYNC

Files checked: 12
Matches: 12
Drifts: 0

Example: Compliance Review

/gov review --all
[CONSTRAINT REVIEW QUEUE]

Compliance Status (SOC 2):

1. CON-20260101-001: "Always encrypt PII at rest"
   Age: 45 days | Status: active
   Compliance: SOC 2 CC6.1
   Violations (90d): 0 | Adoption: 100%
   ✓ Compliant

2. CON-20260115-002: "Always log authentication events"
   Age: 31 days | Status: active
   Compliance: SOC 2 CC6.2
   Violations (90d): 1 | Adoption: 98%
   ⚠ Review violation on 2026-02-01

Summary: 12 constraints | 11 compliant | 1 needs review

Example: Security Audit Preparation

/gov state --summary
[GOVERNANCE STATE]
Updated: 2026-02-15 14:00:00

Audit Readiness:
  Security constraints: 8 active
  Last review: 2026-02-10
  Violations (90d): 2 (both resolved)
  Override rate: 5% (within policy)

Recommendation: Ready for external audit.

Integration

• Layer: Governance
• Depends on: constraint-engine (for constraint data), failure-memory (for observation data)
• Used by: None (top-level governance)

Failure Modes

Next Steps

After invoking this skill:

Workspace Files

This skill reads/writes:

output/
├── governance/
│   ├── state.json           # Central governance state
│   ├── reviews/             # Review records
│   │   └── YYYY-MM-DD.md
│   └── alerts.json          # Pending alerts
└── constraints/
    └── metadata.json        # Constraint metadata (adoption, violations)

agentic/
└── INDEX.md                 # Generated skill index

Security Considerations

What this skill accesses:

• Configuration files in .openclaw/governance.yaml and .claude/governance.yaml
• Constraint data from output/constraints/ (via constraint-engine)
• Observation data from .learnings/ (via failure-memory)
• Its own output directory output/governance/
• Skill index file agentic/INDEX.md

What this skill does NOT access:

• Files outside declared workspace paths
• System environment variables
• Network resources or external APIs

What this skill does NOT do:

• Send data to external services
• Execute arbitrary code
• Modify files outside its workspace

Dependency note: This skill reads data from constraint-engine and failure-memory skill workspaces. Install the full governance stack for complete functionality.

Acceptance Criteria

• [ ] /gov state shows complete governance overview
• [ ] /gov state surfaces alerts for due reviews
• [ ] /gov review lists constraints due for 90-day review
• [ ] /gov review provides clear renewal/retirement options
• [ ] /gov index generates skill index from SKILL.md files
• [ ] /gov verify detects drift between source and compiled
• [ ] /gov migrate handles schema version transitions
• [ ] Adoption metrics tracked and reported
• [ ] Workspace files follow documented structure

*Consolidated from 6 skills as part of agentic skills consolidation (2026-02-15).*