The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from potentially untrusted files within a repository, such as READMEs, contribution guidelines, and existing instruction files (e.g., CLAUDE.md, AGENTS.md). This creates a surface where malicious instructions embedded in these files could influence the agent's behavior during the creation or auditing of the final instructions.
Ingestion points: The skill explicitly reads README*, CONTRIBUTING*, AGENTS.md, CLAUDE.md, GEMINI.md, and .github/copilot-instructions.md (Sections 1 and 2).
Boundary markers: There are no instructions for the agent to use explicit delimiters or "ignore embedded instructions" markers when processing this external content.
Capability inventory: The skill is capable of modifying files (AGENTS.md) and executing shell commands (e.g., npm test, lint) for validation.
Sanitization: While the skill includes a safety note to avoid including secrets, it does not specify methods for sanitizing or escaping the content read from the repository files before interpolation.
[COMMAND_EXECUTION]: The skill instructs the agent to run commands found in the repository's configuration (like npm test, lint, or typecheck) to validate the documentation. While the skill advises running only "safe, proportionate checks" and avoiding destructive commands, this still involves executing code derived from untrusted repository sources.

agents-md