git-branch-cleanup
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- PROMPT_INJECTION (HIGH): Indirect Prompt Injection (Category 8) risk detected. The skill ingests untrusted metadata (branch names and commit subjects) from the local Git environment to drive logic and shell parameters. 1. Ingestion points: The skill uses
git for-each-refandgit branch -vto read repository data. 2. Boundary markers: None present. 3. Capability inventory: High-privilege destructive Git operations. 4. Sanitization: Minimal; some script examples rely on shell interpretation of branch names. - COMMAND_EXECUTION (HIGH): The skill facilitates the execution of destructive shell commands (
git branch -Dandgit worktree remove --force). These commands can result in permanent data loss and are executed based on inputs derived from the repository metadata.
Recommendations
- AI detected serious security threats
Audit Metadata