The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates reading and rendering content from external databases and views, creating a potential surface for indirect prompt injection where malicious data could influence agent actions.
Ingestion points: External data is ingested via LOCAL_NOTIS_DATABASE_QUERY, LOCAL_NOTIS_DATABASE_GET_DATABASE, and LOCAL_NOTIS_RENDER_VIEW as described in the documentation.
Boundary markers: The skill documentation in references/documents.md explicitly warns the agent to treat the page_context as an 'untrusted reference' and to re-verify state before acting on it.
Capability inventory: The skill provides high-impact capabilities including record mutation (LOCAL_NOTIS_DATABASE_INSERT, LOCAL_NOTIS_DATABASE_UPDATE), database deletion (LOCAL_NOTIS_DATABASE_DELETE_DATABASE), and binary file uploads (LOCAL_NOTIS_DATABASE_UPLOAD_FILE).
Sanitization: The instructions focus on manual verification of data freshness and access rights rather than describing automated sanitization or output encoding for the ingested content.

notis-query