The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local git commands (git status, git diff, git add, git commit, git branch) to manage repository state as its primary function.
[PROMPT_INJECTION]: The skill processes untrusted content from file diffs and commit logs to generate commit messages and determine branch names, creating a surface for indirect prompt injection.
Ingestion points: Reads output from git diff and git log to analyze changes.
Boundary markers: No explicit delimiters or warnings to ignore embedded instructions within the ingested data are present.
Capability inventory: Can stage files (git add), create/switch branches, and perform commits.
Sanitization: Employs a specific exclusion list for sensitive files and uses secure shell formatting (quoted HEREDOCs) for the final commit command.
[SAFE]: Implements a robust exclusion policy that explicitly forbids staging sensitive items such as .env files, .pem keys, and credentials.json files, preventing accidental data exposure.
[SAFE]: Utilizes a secure shell pattern (cat <<'EOF') for the git commit command, which prevents the agent from inadvertently triggering shell expansion or command injection through the generated commit message content.

commit