Developed an end-to-end cleanup feature for stale AI agent git worktrees in the tw93/Mole repository, focusing on resource hygiene and reliability within CI pipelines. The solution leveraged bash scripting, git, and devops practices to safely reclaim disk space by removing unused worktrees while preserving any with uncommitted changes. The workflow introduced opt-in controls, safety nets such as Trash routing and dry-run modes, and performance safeguards like timeout-wrapped git calls. Test coverage was expanded using bats to verify both default skip behavior and selective removal scenarios, ensuring robust and predictable operation aligned with existing cleanup patterns in the codebase.