The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill scans the entire codebase and reads external documentation files, creating a surface where untrusted data could influence agent behavior.
Ingestion points: Scans files including README, package.json, CLAUDE.md, and project-specific context files. It specifically looks for TODO, FIXME, and HACK comments within the code.
Boundary markers: The instructions do not define delimiters or specific boundary markers to isolate potentially malicious instructions embedded in the code comments or documentation.
Capability inventory: The skill itself reads files and generates a ranked list of recommendations. It routes the user to other powerful skills such as build, deploy, secure, and finances based on its findings.
Sanitization: There is no mention of sanitizing or escaping the content read from the project files before it is processed by the agent.

next