The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes proactive instructions that forbid the agent from staging sensitive files, including environment files (.env), private keys (*.key, .pem), and credential stores (.p12, *.jks). This prevents accidental exposure of secrets during the commit process.
[COMMAND_EXECUTION]: The skill uses local git commands (git add and git commit) to manage code changes. Security is maintained through mandatory user confirmation via the AskUserQuestion tool before any file is staged or any commit is finalized.
[PROMPT_INJECTION]: The skill processes untrusted input from the local repository (such as git diffs and plan files) to generate commit messages. While this constitutes an indirect prompt injection surface, the requirement for manual user review and approval of the generated message provides a strong mitigation against malicious instruction execution.

create-commit