The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from research papers, web pages, and code repositories.
Ingestion points: Reads data via hf papers read, web search results, and manual downloads of papers and code archives as described in SKILL.md and references/source-acquisition.md.
Boundary markers: The instructions require the agent to label derivatives (e.g., "Markdown rendering from HF") and maintain a structured source.md record to track artifact provenance and reading coverage.
Capability inventory: Utilizes the hf-cli tool and rg (ripgrep) for searching, and requires filesystem write access to the research/ directory.
Sanitization: The skill includes explicit security checks for archive handling, directing the agent to "reject absolute paths, traversal paths, and escaping links." It also strictly prohibits the execution of downloaded code or the compilation of LaTeX source files.
[COMMAND_EXECUTION]: The skill uses hf-cli and ripgrep (rg) for literature discovery and local index searching. These tools are used appropriately for the skill's primary research purpose.
[EXTERNAL_DOWNLOADS]: Fetches research materials from well-known platforms including Hugging Face, ArXiv, and GitHub. The instructions emphasize verifying file integrity using SHA-256 hashes and checking versions at canonical sources.

literature-research