What problem does it solve?
Large repositories accumulate dead surface over time: files nothing imports, components reachable only from their own tests, orphaned i18n locale keys, stale allowlist entries, and exports no consumer uses. Manually proving something is truly unused is error-prone because consumers hide in build scripts, runtime loaders, published packages, and store manifests that no import grep can see.
Core Features & Use Cases
- Repo-wide dead surface survey: Sweeps the qwen-code codebase for six candidate classes (orphan files, dead components with tests, stale registry rows, orphan locale keys, added-then-removed scaffolding, unused exports) using an eight-step proof protocol covering recency gates, published-surface reachability, full-corpus greps, and hidden-consumer checks.
- Issue-first governance workflow: Files candidates as comments on a tracking issue and only lands deletions after explicit maintainer assent, keeping a tombstone ledger so declined candidates are never re-proposed.
- Guarded landing phase: Re-verifies evidence against current origin/main, deletes the surface plus its tests, snapshots, locale keys, and docs rows, then runs build, typecheck, lint, and targeted vitest gates before opening a single-candidate PR.
- Use Case: A maintainer asks for a quarterly cleanup sweep; the skill surveys packages/cli/src, proves three components have no production consumer, files them on the tracking issue, and later lands the one a maintainer approved as a single focused PR.
Quick Start
Ask the agent to run the find-simplifications survey over the qwen-code repository and file any dead-code candidates it proves on the tracking issue.