sweep
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local shell and git commands to manage workspace hygiene.
- It executes
git worktree pruneandgit branch -Dto reclaim storage and manage repository metadata. - It invokes a bundled local script
scripts/prune-squashed.shto identify and remove squash-merged branches. - [INDIRECT_PROMPT_INJECTION]: The skill processes files within the workspace that could contain untrusted data, creating a potential injection surface.
- Ingestion points: Reads subagent traces in
.agents/scratch/and task dossiers in.agents/tasks/to determine the status of work. - Boundary markers: The skill implements a mandatory "Pre-Mutation Gate" where all proposed deletions are presented to the user in a staged manifest for affirmative authorization.
- Capability inventory: Execution capabilities are limited to file system deletion and local git branch management.
- Sanitization: The skill looks for specific status headers (e.g.,
Status: done) to validate reclamation candidates.
Audit Metadata