new-worktree
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute Git commands, including
git fetch originandgit worktree add, to manage local repository states. - The commands are restricted to a defined scope within
~/githuband include checks to ensure the target is a valid Git repository. - [PROMPT_INJECTION]: The skill uses user-supplied input to generate branch and directory names, creating an indirect prompt injection surface.
- Ingestion points: User-provided strings for branch descriptions (e.g., "fix-auth-crash") in
SKILL.md. - Boundary markers: The instructions include hard boundaries for file paths (
~/github) and repository types. - Capability inventory: The skill uses
git worktree addwhich involves shell execution of the derived branch name. - Sanitization: The skill explicitly directs the agent to "Derive a clean, hyphen-separated branch name," which serves as a mitigation for malformed or malicious input.
Audit Metadata