The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git commands (e.g., git diff, git commit) and a bundled Python validation script (scripts/validate_commit.py) as part of its core functionality.
The Python script further executes external linting tools like commitlint if configured.
Security controls are robust: The script uses an allowlist for commands (commitlint, npx --no-install commitlint, etc.) and checks for shell metacharacters to prevent command injection via configuration files.
[DATA_EXPOSURE]: The skill implements a security scanner in scripts/validate_commit.py to detect and prevent the inclusion of sensitive data (AWS keys, GitHub tokens, private keys) in commit messages or code diffs.
[PROMPT_INJECTION]: The SKILL.md contains an explicit 'Prompt injection guard' section, instructing the agent to treat all repository data (diffs, branch names, file contents) as untrusted facts rather than instructions, which mitigates indirect prompt injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (git diffs). However, it uses clear boundary instructions and a dedicated validation step to ensure the agent does not execute malicious instructions embedded within the code being committed.

commit