WorkerFlow uses a docs-first connector model.
Instead of maintaining a large placeholder connector code catalog, the project keeps:
- executable runtime connector primitives
- a shared connector registry used by runtime and ops checks
- a service API docs index for LLM/agent implementation workflows
- Runtime secret requirements and route mapping:
cloudflare/shared/connector_registry.ts
- Service docs index:
docs/SERVICE_API_INDEX.mddocs/SERVICE_API_INDEX.htmlcloudflare/connector-registry/services.json
- Implementation contract:
docs/CONNECTOR_BUILD_SPEC.md
- Agent execution path:
docs/AI_CONNECTOR_ONRAMP.md
WorkerFlow currently ships executable connector-backed routes for:
- chat webhooks (
chat_notify,incident_create) - Slack (
slack_message) - GitHub (
github_issue_create) - OpenAI (
openai_chat) - Stripe (
stripe_payment_intent_create,stripe_customer_upsert) - Notion (
notion_database_item_create,notion_database_item_get) - HubSpot (
hubspot_contact_upsert,hubspot_deal_upsert) - lead normalization AI key requirements (
lead_normalizer)
cd cloudflare
npm run test:connector-registry
npm run test:service-registry
npm run test:connector-harnessBoth checks are included in npm run release:check.
- add/update a service entry in
cloudflare/connector-registry/services.json - optionally scaffold files with
cd cloudflare && npm run connector:new -- --service "<name>" --route <route_id> - add/update runtime route secret mapping in
cloudflare/shared/connector_registry.ts(if route behavior changes) - implement connector route/primitive using
docs/CONNECTOR_BUILD_SPEC.md - run
npm run release:check
- lower maintenance than large scaffold catalogs
- easier for agents to generate connector implementations from official APIs
- clearer distinction between executable runtime behavior and future service coverage