The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest large amounts of untrusted data from external websites and repositories, which creates a vulnerability surface for indirect prompt injection.
Ingestion points: Content is retrieved from the open web using WebFetch and gh search (referenced in SKILL.md).
Boundary markers: The instructions do not prescribe specific delimiters or "ignore instructions" warnings for the agent to use when processing retrieved content.
Capability inventory: The agent has access to command execution through the gh CLI and filesystem access for writing reports to the ./report/ directory.
Sanitization: There is no mention of sanitizing or validating external content before it is used to inform refined search queries or synthesized into the final report.
[COMMAND_EXECUTION]: The skill utilizes the gh CLI to perform repository and code searches. It explicitly instructs the agent to batch multiple search calls into parallel shell executions, where the arguments are dynamically generated from research topics.
[EXTERNAL_DOWNLOADS]: The methodology involves fetching data from a wide array of external platforms, including academic preprints, code registries, and specialized community forums. These operations target well-known and established services like GitHub, arXiv, and Stack Overflow.

research