The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves information from the web and integrates it into local files, providing a surface for malicious content from external sources to influence the agent.\n
Ingestion points: External content fetched from the internet via sub-agents using WebSearch and WebFetch tools.\n
Boundary markers: The skill does not specify delimiters or instructions to ignore commands within the gathered data during the integration process.\n
Capability inventory: The skill utilizes Read, Edit, Write, Glob, Grep, and Agent tools to modify the local filesystem.\n
Sanitization: Instructions focus on verifying source URLs but do not include sanitization of the findings' text for potential injection attacks before they are written to disk.

deep-research