The Agent Skills Directory

[SAFE]: The skill instructions include explicit safeguards, directing the agent to seek user authorization before performing commits, pushes, or history rewrites.
[INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface as it processes untrusted content from staged diffs and commit history (SKILL.md). While this could contain adversarial instructions embedded in code comments, the risk is localized to the generation of commit messages, and the skill lacks automated network exfiltration capabilities.
[DYNAMIC_EXECUTION]: The skill instructs the agent to run pre-commit checks existing within the repository (SKILL.md). This is a standard development workflow and does not involve the generation or execution of new, untrusted, or remote code by the skill itself.

conventional-commits