The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted third-party content (e.g., web pages, GitHub issues, pull request bodies) during research tasks.
Ingestion points: SKILL.md and TRADUCAO.pt-BR.md describe the process of reading primary sources and third-party data.
Boundary markers: Present. The instructions explicitly mandate the use of fenced blocks for untrusted data and include a specific warning for the background agent to disregard embedded instructions.
Capability inventory: The skill allows writing findings to Markdown files in the repository (SKILL.md).
Sanitization: The skill provides clear guidance to the agent to treat commands or instructions found in external data as quotes/facts rather than actions to be performed.

research