The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes untrusted data from external sources, creating a vulnerability to indirect prompt injection where malicious instructions in the source content could attempt to influence agent behavior.\n
Ingestion points: Phase 1 fetches content from arbitrary URLs (articles, tweets, documents) provided by the user or found via search.\n
Boundary markers: The skill lacks explicit delimiters or instructions for the agent to treat fetched content as untrusted data or to ignore embedded instructions.\n
Capability inventory: The skill uses put_page, file_upload, and add_timeline_entry to modify the local knowledge base within the people/, concepts/, and sources/ directories.\n
Sanitization: There is no mention of sanitization or validation of the fetched content before it is processed and written to the brain.

idea-ingest