The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to prevent shallow overrides. It explicitly states that assertions like "I understand" or "just do it" are never sufficient to bypass the verification process. This acts as a defensive measure against accidental or malicious prompts intended to skip safety checks.
[COMMAND_EXECUTION]: The skill itself does not execute arbitrary shell commands. It defines a logic flow for evaluating when other actions (like commits or migrations) should be permitted. It specifically excludes read-only and local tasks from gating to maintain a high signal-to-noise ratio for security.
[DATA_EXFILTRATION]: No network exfiltration or unauthorized file access patterns were detected. The skill's rules involve 'gating' changes to sensitive data such as secrets, environment variables, and IAM permissions, which is a protective behavior.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download or execution of remote scripts. The skill is standalone and contains no external dependencies in its manifest.
[CREDENTIALS_UNSAFE]: No hardcoded secrets or credentials were found. The skill instructions prioritize the protection of secrets by ensuring that any changes to them are subject to high-depth comprehension checks.

The Standard Comprehension Gate