The Agent Skills Directory

[SAFE]: The skill documents standard Git practices and provides benign command examples for common development tasks.
[DATA_EXPOSURE]: The skill explicitly instructs the user to never commit sensitive files such as .env files, credentials, or API keys, promoting good security hygiene.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface where untrusted repository data (via git status or git diff in SKILL.md) could influence agent actions like generating commit messages. However, this is the primary purpose of the skill. Mitigation is present through explicit instructions to review changes, check for sensitive files, and selectively stage files before committing. The skill also suggests using isolated contexts for delegation.

git-workflow