The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands and the Git CLI to perform its core functions.
Evidence: Use of git status, git diff, git commit, and git push within SKILL.md.
Evidence: Use of bash to dynamically resolve the plugin root path.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (the Git diff) which could contain malicious instructions designed to influence the commit message or behavior.
Ingestion points: git diff --staged and git diff (SKILL.md).
Boundary markers: The skill includes explicit instructions to not trust generic 'commit all' requests as permission to bypass security filters.
Capability inventory: Subprocess calls for git commit, git push, and various bash commands for repository state management (SKILL.md).
Sanitization: Implements a specific secret detection command in references/secret-detection.md that uses grep to filter for keys, tokens, and credentials, as well as a blacklist for sensitive file patterns like .env and *.pem.
[DATA_EXPOSURE]: The skill is designed to prevent data exposure by implementing proactive scanning.
Mitigation: Explicitly forbids pasting secrets, passwords, or tokens in the chat, requiring the use of redacted placeholders (path:line (value omitted)).

git-commit