Guardian icon

Guardian

Policy-based safety gate for AI agents and workflows. Check actions against Guardian policies, enforce previously-evaluated intents, or gate AI agent tool calls. Returns Allowed (including OBSERVED advisory decisions), Denied, or Needs Approval.

Actions3

Enforce

AI-generated

Summary

Verifies a Guardian intent by its run ID and routes it according to its current decision. With the recommended “Verify and Claim Once” operation, it also atomically claims execution for an allowed intent using the payload stored by Guardian.

Inputs

  • Operation — Choose “Verify and Claim Once” (default) to claim an allowed intent, or “Verify Status Only” to check its status without claiming execution.
  • Intent Run ID (required) — The intentRunId returned by Guardian Agent Gate.
  • Expected Action Type — When provided, require the intent’s action type to match; a mismatch routes the item to Denied.
  • Expected Project Slug — When provided, require the intent’s project slug to match; a mismatch routes the item to Denied.
  • Expected Idempotency Key — When provided, require the intent’s idempotency key to match; a mismatch routes the item to Denied.
  • Test Mode — When enabled, sends the Guardian test mode header with the verification and, if claiming, the execution request.

Output shape

Items are routed to three outputs: Allowed, Denied, and Needs Approval.

Each output item combines the incoming JSON with Guardian’s response and adds _guardian metadata describing the decision, status, verification, and execution claim. In claim mode, an allowed intent must include a valid stored payload; Guardian then attempts to claim it once. A claim already held by this workflow execution is routed to Allowed with a warning not to repeat the action; a claim held by another execution is routed to Denied. Status-only mode does not claim execution. OBSERVED decisions route to Allowed as advisory results.

Links

Discussion