The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the web using search and crawling tools, which creates an attack surface for indirect prompt injection.
Ingestion points: External content is pulled into the agent's context through firecrawl_scrape, firecrawl_crawl, and crawling_exa tools in SKILL.md.
Boundary markers: The skill includes a robust "Untrusted Sources" section that explicitly warns the agent to treat all fetched content as data for citation, ignore instructions found in sources, and flag manipulation attempts.
Capability inventory: The skill uses network-based MCP tools for data retrieval and provides functionality to save synthesized reports to local files.
Sanitization: The skill relies on clear instructional boundaries to separate untrusted source content from the agent's execution logic.

deep-research