The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large amounts of untrusted content from the open web, including developer documentation, search snippets, research papers, and full-page scrapes.
Ingestion points: Untrusted data enters the agent's context through tools such as tavily_search, firecrawl_scrape, firecrawl_research_read_paper, and context7__query-docs (SKILL.md, references/investigate.md).
Boundary markers: The skill relies on natural language instructions in its Guardrails section to warn the agent that returned pages are data, not instructions, but it does not specify programmatic delimiters (e.g., XML tags or unique markers) for the interpolated content (SKILL.md).
Capability inventory: The agent has the ability to search, scrape, and interact with live websites (including form submission and clicking) via the firecrawl_interact toolset (SKILL.md).
Sanitization: The instructions do not define specific sanitization, filtering, or escaping logic for the retrieved external content before processing.
[EXTERNAL_DOWNLOADS]: The skill provides installation instructions and documentation links for official MCP servers hosted by well-known technology providers, including Upstash (Context7), Tavily, and Firecrawl (README.md, references/costs-and-limits.md).

research-stack