The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various git commands to gather context about the repository and its history.
Evidence: Includes git symbolic-ref, git merge-base, git rev-parse, git log, git diff, and git status in SKILL.md.
Context: The commands are used to identify relevant code changes and are constrained to read-only metadata and diff operations.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing external data from the git repository.
Ingestion points: Untrusted data is ingested from commit messages via git log and code changes via git diff and file reads in SKILL.md.
Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions for the content being analyzed, though it mandates a specific output structure.
Capability inventory: The skill's capabilities are restricted to read-only git operations and file reading. It is explicitly prohibited from writing to files, staging, or committing (Never edit files, stage, or commit).
Sanitization: No validation or sanitization is applied to the content of the diffs or commit messages before processing.

debrief