The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git, gitleaks, and GitHub CLI commands using subprocess argument lists, which prevents shell injection. All user-supplied inputs, including commit SHAs and branch names, are strictly validated against cryptographic hash formats and naming rules prior to execution.\n- [EXTERNAL_DOWNLOADS]: The skill relies on industry-standard, trusted tools such as gitleaks and git-filter-repo. Python dependencies are restricted to the well-known gitpython library, with no execution of unverified remote scripts.\n- [PROMPT_INJECTION]: The static analysis warning regarding 'unrestricted mode' is a false positive; the identified text in references/incident-lessons.md actually provides guidance on enforcing security hooks and avoiding dangerous bypass flags.\n- [DATA_EXFILTRATION]: Analysis found no unauthorized network communication. All remote operations are conducted through authenticated Git and GitHub CLI channels for the purpose of repository management.\n- [DYNAMIC_EXECUTION]: A security-hardened 'frozen transport' mechanism is used in the safe push script to isolate the Git environment from external configuration changes, mitigating potential environment-based attacks during history rewriting.

github-sensitive-data-cleanup