The Agent Skills Directory

[SAFE]: The skill defines a procedural workflow for auditing internal repository documentation against implementation code using standard development tools.
[COMMAND_EXECUTION]: The skill instructs the agent to use git diff to compare the repository's state against a base reference. This is a standard operation for repository management and does not involve elevated privileges or suspicious command chaining.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a theoretical attack surface for indirect prompt injection as it ingests and processes untrusted code diffs and documentation.
Ingestion points: Git diff output containing potentially untrusted code and context modifications (SKILL.md, Step 1).
Boundary markers: The skill lacks explicit syntactic delimiters but provides logical instructions for a sub-agent to verify data ('Judge every added line against the code').
Capability inventory: The skill uses file system read access and git command execution to perform its audit.
Sanitization: No explicit sanitization or escaping of diff content is mentioned.
Note: The exposure is inherent to the tool's primary purpose of auditing repository content and is mitigated by the requirement for judgment and cross-referencing against existing files.

audit-project-context