The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates git operations including staging changes and committing files. These are standard development workflows requested by the user and are mediated by the agent's bash tool, which requires user approval before execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes project files and conversation history to generate commit bodies. It uses specialized local agents (weed-comments and weed-prose) to audit this content. While it ingests external data (diffs and issue comments), it implements a multi-step human-in-the-loop review process where the user must approve the final commit message in the bash tool request.
[SAFE_PRACTICES]: The skill encourages atomic commits and explicitly instructs the agent to separate refactoring from behavior changes, which is a security and stability best practice. It also mandates that specific rationale be moved from code comments (where it might be misfiled or exposed) into commit history.

commit