branch-cleanup
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to use standard Git commands for repository maintenance, including
git fetch --prune,git branch -vv,git worktree list,git worktree remove, andgit pushwith--force-with-lease. These operations are well-scoped to the stated purpose of branch and worktree management. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data originating from the Git repository environment, such as branch names, commit metadata, and pull request statuses from a forge.
- Ingestion points: Output from Git CLI commands and metadata from remote Git forges (e.g., PR state and head SHA).
- Boundary markers: The guidance includes extensive 'Deletion Gates' and 'Stop If' conditions that act as logical boundaries, requiring the agent to verify ownership and merge status before taking action.
- Capability inventory: The skill has the capability to delete local directories (worktrees), delete local branches, and modify remote repository state via Git push.
- Sanitization: The instructions emphasize using 'cheapest authoritative evidence' like PR status and explicit identity rules (
<account-id>/<work-unit>/<description>) to validate actions, reducing the risk of being misled by attacker-controlled branch names or content.
Audit Metadata