The Agent Skills Directory

[COMMAND_EXECUTION]: The skill runs the command git diff --cached to access the code changes currently staged in the repository. This is an expected and safe behavior for its core functionality.\n- [PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by processing external data from git diffs.\n
Ingestion points: Diff data is ingested from the local file system using git commands.\n
Boundary markers: Absent; the content is processed without specific delimiters.\n
Capability inventory: The skill can only read the diff and generate a response, with no access to external networks or writable files.\n
Sanitization: No content filtering is performed on the diff data.

commit-message