@toolmark/lint

Static lint for Toolmark tool manifests: the toolmark lint CLI (and a programmatic lint()) checks what makes agents fail — bad names, missing or short descriptions, invalid input schemas, LLM-name collisions, too many tools per page, consequential actions without a hint. It reads manifests (a JSON file, or a live page's test hook through Playwright), not source code, so it works for any stack.

pnpm add -D @toolmark/lint

ESM only; Node ≥ 22.12. @playwright/test ≥ 1.63 is an optional peer (only for --url, which also needs Chromium: pnpm exec playwright install chromium).

pnpm exec toolmark lint --url http://localhost:5173/
pnpm exec toolmark lint --manifest tools.json --format json
# without installing:
npx -p @toolmark/lint toolmark lint --manifest tools.json

Usage: toolmark lint [options]

  --manifest <file>        Lint a manifest file (repeatable)
  --url <url>              Lint tools collected from a live page (repeatable)
  --storage-state <file>   Playwright storage state for --url
  --judge <spec>           Load a judge module (repeatable)
  --budget <n>             Tool-budget threshold (default: 40)
  --format <pretty|json>   Output format (default: pretty)

At least one --manifest or --url is required. Manifests (files and --url pages alike) are validated against @toolmark/lint/manifest.schema.json.

| Exit | Meaning | | ---- | ----------------------------------------------------------------------------------------- | | 0 | No error findings | | 1 | At least one error finding | | 2 | Usage or runtime failure (bad arguments, no input, invalid manifest, missing Playwright…) |

--judge <spec> loads and runs code: the named module (a path, or a package resolved from the current directory) is imported and executed as part of the lint run. Only pass modules you trust. @toolmark/judge-typesafe is the optional TypeSafe judge.

Guide (manifest shapes, rules, judges): docs/guides/lint.md.

Docs: Lint guide · API reference.

License

MIT