ClawGuard Governance Skill

ClawGuard enforces a universal audit and approval layer for any potentially destructive or high-impact change. It persists all intents to clawguard.db and automatically blocks actions when risk_level >= 4.

Tools

• evaluate_action(action_type, justification, risk_level)

  • Logs the requested action to the audit_ledger table and returns { allowed: boolean, message: string }.
  • Risk levels:
    • 1 = negligible (read-only, cosmetic)
    • 2 = low (single-record updates, reversible)
    • 3 = medium (bulk updates, config tweaks)
    • 4 = high (privileged file writes, schema changes, secrets)
    • 5 = critical (system-wide deletions, irreversible ops)
  • Any level >= 4 is blocked automatically—handle accordingly.

• get_audit_report(limit)

  • Fetches the most recent limit rows (default 5) from audit_ledger, ordered by newest first.

Workflow

1. Classify the action — Determine action_type, craft a concise justification, and score risk_level.
2. Call evaluate_action BEFORE executing any SQL statement, file mutation, or write-capable API request. (See GUIDANCE.md.)
3. Honor the result — If allowed is false, stop immediately and surface the block reason.
4. Execute the operation only after an approved response.
5. Audit as needed with get_audit_report(limit) when preparing reports or debugging governance outcomes.

Additional Notes

• audit_ledger schema: (id INTEGER PK, action_type TEXT, justification TEXT, risk_level INTEGER, status TEXT, ts DATETIME DEFAULT CURRENT_TIMESTAMP).
• The ledger is persistent; repeated approvals accumulate chronological history.
• Keep justifications specific (who/what/why) to maintain a high-quality audit trail.