s2g-workflow-engine（s2g 工作流引擎）

S2G Integration

OpenClaw is the orchestrator. S2G is the toolbox. Connect to an S2G workflow via WebSocket, auto-discover all nodes, and execute them as tools.

OpenClaw ──WS──▶ S2G (wss://s2g.run/api/openclaw/ws/{nodeId})
                   ├── PasswordGenerator
                   ├── HashGenerator
                   ├── DateMath
                   ├── SqlServer
                   ├── Knowledge Base
                   └── ... 200+ node types

The OpenClaw Node in S2G

The OpenClaw node is a built-in S2G node type (category: AI) that acts as a bidirectional bridge between OpenClaw agents and S2G workflows. It appears in the S2G node catalog as "OpenClaw Agent" and serves two roles:

1. Bridge Endpoint — Provides a WebSocket endpoint (wss://s2g.run/api/openclaw/ws/{nodeId}) that OpenClaw agents connect to. Once connected, the agent can execute any sibling node in the workflow.

1. Data Forwarder — Pushes upstream workflow data to connected OpenClaw agents via Input Forwarding. When the workflow triggers (e.g., from an HTTP webhook, scheduler, or another node), mapped fields are sent to all connected agents as {"type":"data","data":{...}}.

Setting Up the OpenClaw Node

1. Open the S2G designer at s2g.run
2. Create or open a workflow
3. Add an OpenClaw node from the AI category in the node palette
4. Add tool nodes — any nodes you want the agent to access (e.g., PasswordGenerator, HashGenerator, DateMath, SqlServer, Knowledge Base)
5. Connect the OpenClaw node to the tool nodes (the connection wiring tells S2G which nodes to expose)
6. Configure the OpenClaw node (click to open properties):

- Auth Secret (optional) — If set, the bridge must send {"type":"auth","secret":"..."} as its first message - Per-Request Timeout — Maximum time per individual node execution request - Input Forwarding — Map upstream node outputs to push data to connected agents - Manual Payload — Send ad-hoc JSON payloads to connected agents for testing

1. Start the workflow (▶ button or API: POST https://s2g.run/api/v1/workflows/{id}/start)
2. Copy the Node ID (UUID) from the node properties — this is needed for the bridge connection URL

Live View

The OpenClaw node properties panel includes a Live View that shows:

• Connected clients count
• Real-time execute/result message flow
• Data push events
• Connection/disconnection events

Prerequisites

1. S2G account at s2g.run (or self-hosted instance)
2. A workflow with an OpenClaw node and tool nodes connected to it
3. Workflow running — the WebSocket endpoint only accepts connections while the workflow is active
4. Node.js with ws module: npm install ws

Quick Start

1. Install the bridge

# Copy bridge script to workspace
cp scripts/s2g-bridge.js ~/.openclaw/workspace/s2g-bridge.js
npm install ws

2. Get the OpenClaw node ID

In the S2G designer at s2g.run, click the OpenClaw node → copy the Node ID (UUID) from properties.

3. Start the bridge

# Connect to public S2G
node s2g-bridge.js --s2g wss://s2g.run --node-id YOUR_NODE_UUID [--port 18792] [--secret SECRET]

# Or with environment variables:
S2G_WS_HOST=wss://s2g.run S2G_NODE_ID=abc-123 node s2g-bridge.js

# Self-hosted S2G instance (development)
node s2g-bridge.js --s2g ws://YOUR_HOST:5184 --node-id YOUR_NODE_UUID

4. Verify

curl http://localhost:18792/health
# {"healthy":true,"uptime":42.5}

curl http://localhost:18792/nodes
# Lists all discovered nodes with names, types, and IDs

Bridge HTTP API

The bridge exposes a local HTTP API on port 18792 (configurable via --port):

Executing Nodes

By name (recommended)

curl -X POST http://localhost:18792/execute/PasswordGenerator \
  -H "Content-Type: application/json" \
  -d '{"params": {"length": "20", "mode": "strong"}}'

By node ID

curl -X POST http://localhost:18792/execute \
  -H "Content-Type: application/json" \
  -d '{"nodeId": "uuid-here", "params": {"length": "20"}}'

Response

{
  "success": true,
  "output": {
    "Password": "xK9!mN...",
    "Strength": "Very Strong",
    "_TriggeredTags": "[\"success\"]"
  }
}

Node Schema Discovery

Before using an unfamiliar node, check its exact parameter names via the S2G Catalog API:

# Get schema for any node type
curl -H "X-API-Key: $KEY" "https://s2g.run/api/v1/catalog/nodes/Custom_Base64/schema"

# List all available node categories
curl -H "X-API-Key: $KEY" "https://s2g.run/api/v1/catalog/categories"

# List all nodes in a category
curl -H "X-API-Key: $KEY" "https://s2g.run/api/v1/catalog/categories/AI/nodes"

The fieldName in inputFields is the exact key to use in params. Case-sensitive.

Common Node Types

Data Generation

• PasswordGenerator — length, mode (strong/pronounceable/passphrase/PIN), count
• UuidGenerator — count, format
• HashGenerator — text, algorithm (md5/sha1/sha256)
• LoremIpsum — count, unit (paragraphs/sentences/words)

Data Operations

• DateMath — operation (add/subtract/difference), date1, date2, days, hours...
• ExpressionEvaluator — expression, variables, precision
• CronParser — expression, count, fromDate
• TimezoneConverter — datetime, fromTimezone, toTimezone
• JsonPathQuery — json, path, returnFirst

Text & Encoding

• Base64 — inputText, mode (encode/decode)
• UrlEncode — inputText, mode (encode/decode)
• CaseConverter — text, targetCase (camelCase/PascalCase/snake_case/kebab-case)
• MarkdownToHtml — markdown, addWrapper
• TextDiff — oldText, newText

Format Conversion

• XmlToJson — xmlInput
• JsonToXml — jsonInput, rootElement
• CsvJson — inputText, mode (toJson/toCsv), delimiter
• JwtDecoder — token

AI Nodes

• OpenAI — GPT models for text generation
• Anthropic — Claude models
• Gemini — Google AI models
• DeepSeek / DeepSeek Agent — DeepSeek models with agent capabilities
• Orchestrator — Multi-agent orchestration across AI providers

Knowledge Base

• Knowledge — Graph-based knowledge store with 11 operations: Search, GetEntity, AddEntity, UpdateEntity, DeleteEntity, AddRelation, RemoveRelation, GetRelations, GetGraph, ListEntities, ListTypes

Vector Storage

• VectorDb — Vector store for RAG: ingest documents, semantic search
• VectorClient — Query vector stores from workflows

Database

• SqlServer — Execute SQL queries against SQL Server
• Postgresql — Execute PostgreSQL queries
• MongoDB — MongoDB document operations

Data Push (S2G → OpenClaw)

S2G can push data to connected OpenClaw agents in two ways:

1. Input Forwarding — Configure in the OpenClaw node properties. Map upstream node output fields to keys that get forwarded to all connected agents. When the workflow triggers (e.g., webhook receives data → processes it → OpenClaw node forwards results), agents receive {"type":"data","data":{...}}.

1. Manual Payload — In the OpenClaw node properties, use the Manual Payload editor to send ad-hoc JSON to all connected agents. Useful for testing and debugging.

Deployment & Operations

For running as a service (systemd/pm2), connection lifecycle, auto-reconnect behavior, monitoring, handling S2G/OpenClaw restarts, security, and multi-bridge setups: see references/operations.md.

Key points:

• Bridge auto-reconnects on disconnect (5s delay)
• Bridge is stateless — just needs the WebSocket URL and node ID
• Sends keepalive pings every 30s to detect dead connections
• 409 = workflow not running — enable Auto-Start (🚀) in S2G or start via API: POST https://s2g.run/api/v1/workflows/{id}/start
• Logs to logs/s2g-bridge.log with 5MB rotation

S2G REST API

Full platform API at https://s2g.run/api/v1/ covering workflows, catalog, knowledge base, AI assistant, connections, usage, and logs: see references/api.md.

Key capabilities:

• Workflows — Create, start/stop, add/remove nodes and connections: https://s2g.run/api/v1/workflows
• Catalog — Discover all 200+ node types and their schemas: https://s2g.run/api/v1/catalog/nodes
• Knowledge Base — Graph-based knowledge store with search, entities, and relations
• AI Assistant — Generate workflows from natural language: POST https://s2g.run/api/v1/ai/generate
• Connections — Manage OAuth/API key connections: https://s2g.run/api/v1/connections
• Usage & Logs — Monitor quotas and execution logs: https://s2g.run/api/v1/usage

WebSocket Protocol

For raw protocol details (message types, auth handshake, data push framing): see references/protocol.md.

Connection URL: wss://s2g.run/api/openclaw/ws/{nodeId} Health check: GET https://s2g.run/api/openclaw/health

Tips

• No manual wiring needed. All sibling nodes connected to the OpenClaw node are auto-discovered on connect.
• Param names are case-sensitive. Always verify with the catalog schema API.
• _TriggeredTags in output indicates which connection tag fired (success/error).
• 409 on connect means workflow isn't running. Start it at s2g.run or via API.
• Bridge auto-reconnects on disconnect with 5s delay — no manual intervention needed.