The Agent Skills Directory

[SAFE]: The skill implements a secure git workflow with built-in protections against common pitfalls.\n- [SAFE]: A comprehensive blocklist prevents accidental staging of sensitive files, including environment variables (.env*), credentials, private keys (*.key, *.pem), and local settings.\n- [SAFE]: The execution of shell commands is handled securely. AI-generated content (commit messages and summaries) is passed to git via single-quoted heredocs (<<'EOF'), which effectively prevents command injection by disabling shell expansion.\n- [COMMAND_EXECUTION]: The skill uses standard git operations (status, add, commit, notes, push) and local file cleanup (rm) within the Bash tool to perform its tasks.\n- [DATA_EXFILTRATION]: The skill pushes git notes to the repository's origin remote. This is standard for its purpose and uses the user's established git configuration.\n- [PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface where untrusted data (file names from the local filesystem) is interpolated into a shell command.\n
Ingestion points: File names are retrieved via git status --porcelain in Step 1.\n
Boundary markers: None used for the git add <files> argument list; however, heredocs are used for the commit message and summary.\n
Capability inventory: Full Bash execution capability is used for git operations.\n
Sanitization: The skill includes a blocklist for sensitive file names and explicitly instructs the agent to avoid shell expansion, which serves as a mitigation for processing untrusted file names.

memento-commit