Skip to main content

Scoped permissions

Policies tie a principal (user or agent) to path patterns and permissions. Use narrow patterns and only the permissions needed.

Path patterns​

  • ** — All paths in the vault (read or write depending on permissions).
  • api-keys/* — Only paths under api-keys/ (e.g. api-keys/openai, api-keys/stripe).
  • prod/db — Exact path prod/db only.

Patterns use glob rules: * matches one segment, ** matches zero or more segments. Give agents the smallest set of paths they need.

Permissions​

  • read — Can GET secret value and list (metadata).
  • write — Can PUT (create/update) secrets at matching paths.
  • delete — Can DELETE secrets at matching paths.
  • rotate — Can trigger server-side secret rotation.
  • list — Can list secret metadata.

Grant only read unless the agent must create or update secrets. Note that write does not include delete — grant delete explicitly if needed.

Example: agent for one key​

Policy: principal_type agent, principal_id <agent_uuid>, secret_path_pattern api-keys/openai, permissions ["read"]. The agent can only read that one path.

Example: agent for all keys, read-only​

secret_path_pattern api-keys/*, permissions ["read"]. The agent cannot write or delete.

Last updated on