The Agent Skills Directory

[SAFE]: The skill is entirely instructional and does not contain any executable scripts, binary files, or external package dependencies.
[DATA_EXPOSURE]: The instructions include specific safety guardrails directing the agent to mention sensitive configuration keys (like environment variables) without exposing their actual secret values.
[COMMAND_EXECUTION]: While the skill mentions using git commands like git diff and git status, it explicitly states that the skill's purpose is limited to drafting text. It delegates actual repository modifications (staging, committing, pushing) to the surrounding workflow or user oversight.
[PROMPT_INJECTION]: The skill implements a verification step requiring the agent to trace every claim in the commit message back to a specific code hunk in the diff, which serves as a defense against hallucination or indirect prompt injection from malicious comments in the code being analyzed.

commit-message