knip — Claude Code skill: find dead code and unused dependencies

What knip does

The skill gives the agent an order of operations: understand the project first, clean second. It reads package.json, looks for an existing knip config (knip.json, knip.jsonc or a knip key in package.json), then runs npx knip. Configuration hints at the top of the output come first, because they cut down false positives. Run, adjust config, run again — that loop repeats until the hints stop appearing.

Reported issues come next, in a fixed order: unused files, then dependencies and devDependencies, then exports, then types. Knip gets re-run after each batch, since deleting files tends to expose exports and dependencies that just lost their last consumer. For production-only checks there is npx knip --production, which leaves out tests and configs. Safe fixes go through npx knip --fix; file deletion needs npx knip --fix --allow-remove-files.

One limit matters: knip does not touch unused imports or variables inside a file. That is a linter's territory — ESLint or Biome. The agent deletes some findings on its own and asks before the risky ones: possible entry points or dynamically imported files, exports from index.ts and lib/, dependencies used through a CLI or as peers. Exit code 2 sends the skill looking for a missing config and creating knip.json at the project root. Monorepos may need the --workspace flag.

Node.js is required.