The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external repository data, such as AGENTS.md, project manifests, and GitHub Actions logs, to determine workflow requirements. While this introduces a potential ingestion point for untrusted instructions, the skill mitigates this by mandating the use of predefined repository PR templates and requiring evidence-based validation for all automated PR content.\n- [COMMAND_EXECUTION]: The skill automates complex Git and GitHub workflows using local CLI tools. It avoids hardcoded commands by dynamically discovering syntax through help flags, ensuring that it only executes supported and verified operations within the user's environment.\n- [DATA_EXFILTRATION]: The skill incorporates explicit safeguards against the exposure of sensitive data. It identifies and protects files like .ssh/id_rsa and .env during worktree operations, and it limits network communication to well-known service endpoints like GitHub for PR and release management.

git-workstreams