CrewSwarm Plugin for OpenClaw

Connects your OpenClaw agents to a local CrewSwarm multi-agent crew.

Your OpenClaw agents gain three new tools — crewswarm_dispatch, crewswarm_status, and crewswarm_agents — plus a /crewswarm slash command and Gateway RPC methods. No LLM credentials are shared — only a single auth token.

What it does

| Surface | Description | |---|---| | crewswarm_dispatch | Agent tool — dispatch a task to any crew agent and block until done | | crewswarm_status | Agent tool — poll status of a task by taskId | | crewswarm_agents | Agent tool — list available agents | | /crewswarm <agent> <task> | Slash command from any channel | | crewswarm.dispatch | Gateway RPC | | crewswarm.status | Gateway RPC | | crewswarm.agents | Gateway RPC |

Requirements

• CrewSwarm running locally (npm run restart-all)
• crew-lead reachable at http://127.0.0.1:5010 (default)
• Your RT auth token from ~/.crewswarm/config.json → rt.authToken

Install

# From the CrewSwarm repo root:
openclaw plugins install ./contrib/openclaw-plugin

Or link for development (no copy, reflects edits immediately):

openclaw plugins install -l ./contrib/openclaw-plugin

Then restart the OpenClaw Gateway:

openclaw restart

Configure

Add to your openclaw.json:

{
  "plugins": {
    "entries": {
      "crewswarm": {
        "enabled": true,
        "config": {
          "url":   "http://127.0.0.1:5010",
          "token": "<your RT auth token>"
        }
      }
    }
  }
}

Find your token:

cat ~/.crewswarm/config.json | python3 -c "import json,sys; print(json.load(sys.stdin)['rt']['authToken'])"

Optional config:

| Key | Default | Description | |---|---|---| | url | http://127.0.0.1:5010 | crew-lead base URL | | token | (required) | RT auth token | | pollIntervalMs | 4000 | Status poll frequency | | pollTimeoutMs | 300000 | Max wait time (5 min) |

Usage

From an OpenClaw agent conversation

Your OpenClaw agent will automatically call crewswarm_dispatch when it makes sense:

"Use crew-coder to write a login endpoint with JWT auth"

Or explicitly:

"Call crewswarm_dispatch with agent=crew-qa to audit my last change"

Slash command (any channel — Telegram, WhatsApp, etc.)

/crewswarm crew-coder write /tmp/hello.js — a 10-line express hello world
/crewswarm crew-qa audit the last PR changes
/crewswarm crew-pm create a roadmap for the auth feature
/crewswarm                   ← lists available agents

Gateway RPC (from scripts or other tools)

# Dispatch
openclaw rpc crewswarm.dispatch '{"agent":"crew-copywriter","task":"Write a tagline for CrewSwarm"}'

# Poll status
openclaw rpc crewswarm.status '{"taskId":"<uuid>"}'

# List agents
openclaw rpc crewswarm.agents

Direct HTTP (no OpenClaw needed)

TOKEN="<your RT token>"

# List agents
curl -H "Authorization: Bearer $TOKEN" http://127.0.0.1:5010/api/agents

# Dispatch
curl -X POST http://127.0.0.1:5010/api/dispatch \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"agent":"crew-coder","task":"write hello.js"}'

# Poll
curl -H "Authorization: Bearer $TOKEN" http://127.0.0.1:5010/api/status/<taskId>

Available agents (default crew)

| Agent | Role | |---|---| | crew-coder | Full-stack coding | | crew-coder-front | Frontend specialist | | crew-coder-back | Backend specialist | | crew-frontend | UI/CSS polish | | crew-qa | Testing & audit | | crew-fixer | Bug fixing | | crew-pm | Planning & roadmaps | | crew-security | Security review | | crew-copywriter | Writing & docs | | crew-github | Git & PRs | | crew-main | General purpose | | crew-mega | Heavy reasoning tasks | | crew-researcher | Web research | | crew-architect | System design | | crew-ml | ML/data tasks | | crew-orchestrator | Wave orchestration | | crew-seo | SEO optimization | | crew-loco | Web search/fetch | | crew-judge | Task evaluation | | crew-telegram | Telegram integration | | crew-whatsapp | WhatsApp integration |

How it works

OpenClaw agent
  → crewswarm_dispatch tool call
    → POST /api/dispatch (crew-lead HTTP, Bearer token)
      → RT WebSocket bus (port 18889)
        → target agent bridge (gateway-bridge.mjs)
          → LLM call (Mistral / Cerebras / DeepSeek / etc.)
          → tool execution (@@WRITE_FILE, @@RUN_CMD, etc.)
        → task.done event back on RT bus
      → crew-lead stores result
    → GET /api/status/:taskId (polled every 4s)
  ← result returned to OpenClaw agent

No LLM keys cross the boundary — CrewSwarm uses its own provider config. The only shared secret is the RT auth token.

Troubleshooting

"crew-lead unreachable" — run npm run restart-all in the CrewSwarm repo.

401 Unauthorized — token mismatch. Re-copy from ~/.crewswarm/config.json → rt.authToken.

Task never completes — check /tmp/crew-lead.log and the bridge log in /tmp/.

Agent not found — call crewswarm_agents to see the live agent list.