clawhub-importer

Crawl every skill from ClawHub, migrate metadata.openclaw to metadata.strawpot, and publish to StrawHub.

See DESIGN.md for detailed architecture.

Install

pip install -e .

Requires Python 3.10+.

Usage

Dry run (crawl + transform, no publish)

clawhub-importer --dry-run

Dump transformed skills to disk for inspection

clawhub-importer --dump-dir ./output

Publish to StrawHub

export STRAWHUB_URL="https://your-strawhub-instance.dev"
export STRAWHUB_TOKEN="your-api-token"
clawhub-importer --publish

Or pass the target URL explicitly:

clawhub-importer --publish --target https://your-strawhub-instance.dev --token your-api-token

Import specific skills only

clawhub-importer --publish --slugs gridtrx 12306

Force re-import (ignore state)

clawhub-importer --publish --force

All options

clawhub-importer [OPTIONS]

  --target URL         StrawHub base URL (or set STRAWHUB_URL env var)
  --token TOKEN        StrawHub API Bearer token (or set STRAWHUB_TOKEN env var)
  --dry-run            Crawl and transform but don't actually publish
  --publish            Actually publish to StrawHub (requires --token and --target)
  --dump-dir DIR       Directory to dump transformed skills for inspection
  --slugs SLUG [...]   Only process specific skill slugs
  --force              Re-import all skills, ignoring previous state
  --state-file PATH    Path to import state file (default: .clawhub_importer_state.json)
  -v, --verbose        Enable debug logging

Incremental imports

The importer tracks which skills have been imported and at which version in .clawhub_importer_state.json. On subsequent runs, it skips unchanged skills and only downloads new or updated ones. Use --force to re-import everything.

Skills claimed by another user on StrawHub (400 "You do not own this skill") are permanently added to a skipped_slugs list in the state file, so they are never re-downloaded or re-attempted.

CI / GitHub Actions

A daily import workflow runs via .github/workflows/import.yml:

• 06:00 UTC — imports to the preview environment
• 08:00 UTC — imports to production

The workflow uses actions/cache to persist the state file between runs. Manual dispatch is also supported via workflow_dispatch with a target selector (preview or production).

Required repository secrets: STRAWHUB_PREVIEW_URL, STRAWHUB_PREVIEW_TOKEN, STRAWHUB_PROD_URL, STRAWHUB_PROD_TOKEN.

Metadata migration

The importer preserves original metadata.openclaw / metadata.clawdbot and adds metadata.strawpot alongside it:

Before (ClawHub):

metadata: {"openclaw":{"emoji":"...","requires":{"bins":["node"],"env":["API_KEY"]}}}

After (StrawHub):

metadata:
  openclaw:
    emoji: "..."
    requires:
      bins: [node]
      env: [API_KEY]
  strawpot:
    dependencies: []
    tools:
      node:
        description: "Required binary: node"
        install:
          macos: brew install node
          linux: apt install nodejs
          windows: winget install OpenJS.NodeJS

Rate limiting

The importer respects ClawHub's rate limits (120 req/60s for list API, 20 req/60s for downloads) by reading response headers and throttling automatically. Filtering by state before downloading avoids wasting the expensive download quota on unchanged skills.

License

MIT