The Agent Skills Directory

[SAFE]: The skill performs standard version control tasks using the git command-line tool. All operations align with the stated purpose of the skill.
[DATA_EXPOSURE]: The skill includes a specific safety instruction: 'Do NOT include files that may contain secrets (.env, credentials, tokens, etc.)', which helps prevent accidental leakage of sensitive information during the commit process.
[COMMAND_EXECUTION]: Usage of shell commands is scoped to the git binary as specified in the allowed-tools frontmatter, providing a restricted execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes file diffs (external untrusted data) to generate commit messages. While this presents a surface for indirect prompt injection, the skill provides detailed heuristics and multi-step logic ('Task' and 'Commit Message Convention' sections) that guide the agent's reasoning, mitigating the risk of following instructions embedded within code changes.

commit