The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources (ShiftCare MCP tools) which may contain user-generated content or instructions designed to influence agent behavior.
Ingestion points: Untrusted data enters the context through list_complaints, get_complaint, list_staff, list_clients, and list_incidents (SKILL.md).
Capability inventory: The skill possesses significant write capabilities, including create_complaint, update_complaint, and update_complaint_status (SKILL.md).
Boundary markers: The skill includes strong mitigations, specifically a mandatory 'Confirm, then write once' workflow that requires the user to review a literal block of the proposed data before execution.
Sanitization: The instructions require re-verifying the account_id via whoami immediately before any write operation to ensure session consistency and prevent cross-account data leakage.

shiftcare-complaints