AANA Guardrail Skill Bundled Helper Variant

# AANA Guardrail Skill Bundled Helper Variant

Safety Notice

This listing is from the official public ClawHub registry. Review SKILL.md and referenced scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "AANA Guardrail Skill Bundled Helper Variant" with this command: npx skills add mindbomber/aana-guardrail-bundled

AANA Guardrail Skill Bundled Helper Variant

Use this variant only when the user or administrator wants a bundled, inspectable helper inside the skill package.

This package includes a small Python helper at bin/aana_guardrail_check.py. The helper does not implement AANA policy itself. It only sends a minimal review payload to a separately running AANA HTTP bridge on localhost.

Safety Boundary

  • Do not run the helper unless the user explicitly approves using the configured AANA bridge for this task.
  • Do not send payloads to remote hosts. The bundled helper rejects non-localhost URLs.
  • Do not include API keys, bearer tokens, passwords, full payment numbers, unnecessary account records, or unrelated private messages.
  • Prefer redacted summaries over raw sensitive content.
  • Treat AANA recommendations as advisory unless the bridge and policy are trusted and auditable.
  • If the helper or bridge is unavailable, use manual review.

Review Payload

Use a minimal JSON payload:

{
  "adapter_id": "support_reply",
  "request_summary": "draft a refund support reply",
  "candidate_summary": "reply would promise refund eligibility",
  "evidence_summary": ["refund eligibility is unknown"],
  "allowed_actions": ["accept", "revise", "ask", "defer", "refuse"]
}

Helper Behavior

The helper:

  • reads one JSON payload file,
  • validates that it is a JSON object,
  • blocks obvious secret-like keys,
  • sends the payload to a localhost AANA bridge,
  • prints the JSON result,
  • exits nonzero if the result recommends anything other than accept.

The helper does not:

  • install dependencies,
  • execute other scripts,
  • infer checker paths,
  • contact remote hosts,
  • create event files,
  • delete user files.

Decision Rule

  • If recommended_action is accept, proceed.
  • If recommended_action is revise, revise before acting.
  • If recommended_action is ask, ask the user for missing information.
  • If recommended_action is defer, route to a stronger tool or human review.
  • If recommended_action is refuse, pause, explain the reason, and ask for review when the decision affects important work.

If no trusted bridge is configured, do not use this helper.

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Coding

fleet

Multi-agent fleet management for OpenClaw, Claude Code, Codex, Cursor, Windsurf, Gemini CLI, and custom agent stacks. One CLI to monitor, manage, and orchest...

Registry SourceRecently Updated
Coding

Mini Coder Max

Autonomous coding agent that systematically plans, implements, reviews, and delivers high-quality code. Handles tasks of any complexity by following a struct...

Registry SourceRecently Updated
Coding

Google Docs Writer

Read, write, append, clear, and find-replace content in Google Docs via the Docs API. Use when gog CLI is insufficient (gog only supports read-only doc opera...

Registry SourceRecently Updated
Coding

openclaw-threadsctl

Manage Threads accounts, OAuth connect URLs, drafts, and publishing through the local threadsctl CLI. Use when the user wants to post to Threads, create or p...

Registry SourceRecently Updated