{"skill":{"slug":"aana-guardrail","displayName":"AANA Guardrail Skill","summary":"Monitors agent actions for safety and compliance, preventing violations, private data leaks, or irreversible steps by enforcing predefined decision rules.","description":"# AANA Guardrail Skill For OpenClaw-Style Agents\n\nUse this skill when a user asks the agent to produce or execute an action that may violate hard constraints, depend on missing evidence, expose private information, or require a safe correction path.\n\n## When To Call AANA\n\nCall AANA before:\n\n- sending customer-support replies, emails, or messages,\n- using private account, billing, payment, or personal data,\n- making refund, eligibility, booking, purchase, or policy promises,\n- editing, deleting, moving, or publishing user files,\n- committing or publishing code,\n- answering from incomplete evidence, citations, or source notes,\n- doing anything the user cannot easily undo.\n\n## Command\n\nCreate an event JSON file and run:\n\n```powershell\npython scripts/aana_cli.py agent-check --event path/to/agent_event.json\n```\n\n## Event Shape\n\n```json\n{\n  \"event_version\": \"0.1\",\n  \"event_id\": \"unique-id\",\n  \"agent\": \"openclaw\",\n  \"adapter_id\": \"support_reply\",\n  \"user_request\": \"The user's request\",\n  \"candidate_action\": \"The answer or action the agent is about to take\",\n  \"available_evidence\": [\"Only facts actually available to the agent\"],\n  \"allowed_actions\": [\"accept\", \"revise\", \"ask\", \"defer\", \"refuse\"]\n}\n```\n\n## Decision Rule\n\n- If `gate_decision` is `pass` and `recommended_action` is `accept`, proceed.\n- If `recommended_action` is `revise`, use `safe_response` or revise the plan before acting.\n- If `recommended_action` is `ask`, ask the user for the missing information.\n- If `recommended_action` is `defer`, route to a stronger tool, human review, or verified system.\n- If `recommended_action` is `refuse`, do not execute the candidate action.\n\n## Default Adapter Mapping\n\n- Customer support, refunds, private account details: `support_reply`\n- Travel or booking plans: `travel_planning`\n- Food, grocery, allergy, dietary plans: `meal_planning`\n- Research briefs, cited summaries, knowledge synthesis: `research_summary`\n\nIf no adapter fits, scaffold one:\n\n```powershell\npython scripts/aana_cli.py scaffold \"new workflow name\"\n```\n","tags":{"latest":"1.0.0"},"stats":{"comments":0,"downloads":396,"installsAllTime":0,"installsCurrent":0,"stars":1,"versions":2},"createdAt":1777731724288,"updatedAt":1778492830283},"latestVersion":{"version":"1.0.0","createdAt":1777731724288,"changelog":"AANA Guardrail Skill 1.0.0 - Initial Release\n\n- Introduces a guardrail system for OpenClaw-style agents to enforce hard constraints and safe decision-making.\n- Details specific scenarios when to invoke AANA before agent action (e.g., handling sensitive data, customer support, code commits).\n- Provides command-line usage instructions and JSON event schema for agent checks.\n- Outlines decision rules for agent actions based on AANA recommendations.\n- Includes default adapter mappings for common workflows and instructions for scaffolding new adapters.","license":"MIT-0"},"metadata":null,"owner":{"handle":"mindbomber","userId":"s177cynx168hg36ac5nrkb9wdn85z9dq","displayName":"mindbomber","image":"https://avatars.githubusercontent.com/u/42798111?v=4"},"moderation":{"isSuspicious":false,"isMalwareBlocked":false,"verdict":"clean","reasonCodes":["review.llm_review"],"summary":"Review: review.llm_review","engineVersion":"v2.4.24","updatedAt":1780090732857}}