The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the cloud environment (identity names, ARNs, tags, and descriptions) which is then interpolated into generated CLI and Terraform scripts. 1. Ingestion points: Data enters via tools like discovery_search, get_alerts_with_similar_alert_type, and get_asset_by_name. 2. Boundary markers: The instructions explicitly warn to treat environment data as untrusted and to use single-quotes in scripts. 3. Capability inventory: The skill generates shell and Terraform scripts and interacts with the Orca API to update alert statuses; it does not directly execute generated code. 4. Sanitization: The skill mandates single-quoting all interpolated environment values and requires the agent to exclude identities with names containing shell metacharacters or control characters.
[SAFE]: The skill enforces a mandatory confirmation gate for all destructive (delete) actions. It requires restating the impact and displaying the blast radius (via get_linked_entities_mapping) before proceeding with an explicit user approval.
[SAFE]: The skill includes logic to prevent unauthorized actions based on metadata content, explicitly stating that descriptive text in cloud resources (e.g., 'approved for deletion') must not bypass the formal confirmation process.

orca-inactive-identities-cleanup