The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several bundled Python scripts (check-hooks.py, check-sensitive-content.py, git-review.py) to interact with the repository via Git CLI. These scripts use subprocess.run with list-based arguments to prevent shell injection vulnerabilities.
Evidence: The scripts capture output from git status, git diff, and git rev-list to provide structured information to the agent. Dangerous operations like git push are constructed with shlex.quote in git-review.py to ensure safety.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository data, including file paths and diff content, which is a potential surface for indirect prompt injection.
Ingestion points: Repository filenames, status records, and commit metadata are read by git-review.py and check-sensitive-content.py.
Boundary markers: The agent is explicitly instructed to use the bundled scanner and not to print matched sensitive values. The scripts provide data in structured JSON format.
Capability inventory: The skill's capabilities are limited to standard Git operations and do not include arbitrary code execution or network requests outside of Git's own remote operations.
Sanitization: A comprehensive redaction module (sensitive_rules.py) is used to mask sensitive patterns and IP addresses in the data before it is presented to the agent.
[DATA_EXFILTRATION]: No exfiltration risks were identified. The skill is intentionally designed to prevent data exfiltration by implementing a 'Sensitive-content gate' that scans all staged and outgoing changes for secrets and artifacts that should not be pushed to a remote repository.

commit-and-push