Permission rules

Control which actions the agent can perform without asking you. Permission rules let you fine-tune the trade-off between automation and oversight.

How permission rules work

Every action the agent wants to perform goes through the rules engine before executing. Three possible verdicts:

Rule format

Each rule has:

• tool — the action it applies to (or * for everything).
• verdict — allow, deny, or ask.
• when (optional) — a set of conditions that must all match for the rule to fire.

Conditions can match on the kind of action, on a target path, or on any property of the action's arguments via dotted notation (e.g. payload.status). Globbing (*, ?) is supported in condition values.

Example

[
  {"tool": "*", "verdict": "ask"},
  {"tool": "workflow_apply", "verdict": "allow",
   "when": {"kind": "read"}},
  {"tool": "pfm_apply", "verdict": "allow",
   "when": {"kind": "record", "action": "get"}},
  {"tool": "workflow_create_variable", "verdict": "allow"}
]

This configuration:

1. Asks for confirmation on everything by default.
2. Allows the agent to read workflows without asking.
3. Allows the agent to read records without asking.
4. Allows the agent to create variables without asking.

Managing rules

• Admin UI: Setyenv → Agent → Settings → Permission rules.
• REST: GET / PUT /pfw/v1/agent-runtime/permission-rules.