git-workflows
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No security issues detected. The skill is an instructional framework for professional Git repository management, focusing on correctness and safety.\n- [INDIRECT_PROMPT_INJECTION]: The skill handles repository metadata (commit messages, branch names, and status) which represents a potential indirect injection surface.\n
- Ingestion points: Reads repository instructions, status, and commit messages (SKILL.md, references/papers/147-production-grade-git-and-git-flow.md).\n
- Boundary markers: The skill lacks formal string delimiters for external data but includes strict safety 'Stop Conditions' to mitigate the risk of obeying embedded instructions.\n
- Capability inventory: Provides instructions for local Git mutations, ref updates, and worktree edits.\n
- Sanitization: Instructs the agent to treat all repository code as untrusted input and never to print secrets or bypass safety filters.\n- [DATA_EXFILTRATION]: Sensitive file paths and credential management are discussed strictly in a defensive context. The instructions mandate secret revocation and rotation as a primary response to exposure incidents and forbid printing secrets during diagnosis.\n- [COMMAND_EXECUTION]: While the skill involves executing Git commands, it establishes a rigorous 'Authority and Mutation Gate' that requires explicit authorization and target verification for every mutation, particularly for remote or destructive actions.
Audit Metadata