keep-git-work-recoverable
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a defensive tool focused on repository integrity and data safety. It explicitly prohibits autonomous data destruction and emphasizes human-in-the-loop decision making for risky operations.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted output from Git commands. It manages this risk through strict output formatting and status tagging requirements.
- Ingestion points: Output from git commands like status, branch, and fetch (SKILL.md, rules/locate-yourself.md).
- Boundary markers: Defined 'Output contract' for state reporting (SKILL.md).
- Capability inventory: Git repository management and state inspection commands across multiple rules files.
- Sanitization: Requirement to quote tool errors verbatim and tag remote-dependent claims as Observed or Unverified (SKILL.md, rules/stale-refs.md).
Audit Metadata