The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's 'UPDATE' and 'FIND' modes require the agent to search for and read existing experiment documents within the file system, creating a surface for indirect prompt injection.
Ingestion points: Files are read from several project directories, including experiment/, experiments/, discovery/, product/experiments/, and product/discovery/.
Boundary markers: The skill does not define clear delimiters or instructions for the agent to ignore potentially malicious content or nested instructions within the files it retrieves.
Capability inventory: The skill is configured for file reading and text generation. While the documentation describes workflow management, it does not explicitly invoke high-risk tools like shell execution or external network requests within its logic.
Sanitization: There is no mechanism described for validating or sanitizing the content of the experiment files before they are integrated into the agent's context for updates.

experiment-design