The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes dynamic context injection in the SKILL.md file to execute git branch and git status. These commands are used to provide the agent with local repository state and do not involve untrusted input or network operations, representing a benign and safe use of the feature.
[PROMPT_INJECTION]: While the skill processes user-provided markdown files (a surface for indirect prompt injection), it implements significant safeguards. The instructions define a strict read-only role, and the evaluation suite includes tests specifically verifying that the agent rejects direct prompts to delete or modify files, ensuring it maintains a safe posture.
[SAFE]: The skill does not perform any network operations to external domains, does not include hardcoded credentials, and avoids all forms of code obfuscation. Its primary operations are limited to reading local documentation and performing analysis within the agent's context.

audit-derivability