The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external GitHub repositories, which creates a surface for indirect prompt injection attacks. \n
Ingestion points: Content is fetched via gh CLI commands from issues, pull requests, and file contents across all reference files in the references/ directory. \n
Boundary markers: No explicit delimiters are used to separate external content from agent instructions. \n
Capability inventory: The skill can execute shell commands via the gh CLI and write files to the /tmp/ directory (found in references/code.md and references/releases.md). \n
Sanitization: Content is not sanitized before being presented to the agent. \n- [COMMAND_EXECUTION]: The skill relies on the gh CLI for all operations, which involves executing shell commands to interact with GitHub's REST and GraphQL APIs. \n- [EXTERNAL_DOWNLOADS]: The skill includes functionality to download repository contents and release assets from GitHub's servers to the local /tmp/ directory.

readable-github