The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to perform version control operations using shell commands including git commit, git reset, and git revert to manage file states during the improvement loop.
[COMMAND_EXECUTION]: The agent is prompted to execute local evaluation scripts (e.g., eval.sh, eval.py) and standard developer tools such as lighthouse, k6, and wrk to measure metrics.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) as it involves reading and processing the contents of an arbitrary target file (FILE_TO_MODIFY) which could contain adversarial instructions targeting the agent's logic. * Ingestion points: The agent reads the target file in the loop defined in SKILL.md. * Boundary markers: The skill does not provide specific delimiters or 'ignore' instructions for the target file content. * Capability inventory: The agent can modify files and execute arbitrary shell commands or evaluation scripts. * Sanitization: No evidence of content validation or sanitization is implemented before processing the target file.

autoresearch