Update to RUMBO Agent Reliability

{
  "description": "Independently audit an agent task before accepting its final state.",
  "included_files": [],
  "name": "audit-final-state",
  "skill_md_contents": "---\nname: audit-final-state\ndescription: Independently audit an agent task before accepting its final state.\n---\n\n# Audit Final State\n\nAudit:\n1. Workspace identity.\n2. Canonical authority.\n3. Exact current state/head.\n4. Scope and permissions.\n5. Tool, skill, and dependency provenance.\n6. Human approval for consequential actions.\n7. Execution receipt.\n8. Expected postcondition.\n9. Unexpected side effects.\n10. Rollback or recovery state.\n11. Cross-agent handoff.\n12. Drift since start.\n\nVerdict:\n- GO: all required controls evidenced.\n- CONDITIONAL: blocker has owner and next action/date.\n- NO-GO: acceptance would exceed evidence or authorization.\n\nA passing process is not proof of a passing outcome.\n"
}