swarmclaw

この本文は AI(Claude)が読むための原文(英語または中国語)です。日本語訳は順次追加中。

Overview

SwarmClaw is a self-hosted AI agent orchestration dashboard. This skill gives you CLI access to manage agents, tasks, sessions, schedules, and memory across a SwarmClaw instance.

Install the CLI:

npm i -g @swarmclawai/swarmclaw

Authentication uses SWARMCLAW_ACCESS_KEY (preferred) or CLI key flags. Default URL is http://localhost:3456; override with SWARMCLAW_URL, --url (legacy commands), or --base-url (API-mapped commands).

Use machine-readable output when parsing command results:

--raw for legacy commands
--json for API-mapped commands

Setup

Install: npm i -g @swarmclawai/swarmclaw
Set env var: export SWARMCLAW_ACCESS_KEY=<your-key> (shown in terminal on first run)
Full docs: https://swarmclaw.ai/docs — CLI reference: https://github.com/swarmclawai/swarmclaw

Multi-Gateway Fleet Management

SwarmClaw can connect to multiple OpenClaw gateways at the same time. Each SwarmClaw agent can target a different gateway, including one local and several remote instances. This skill’s commands operate through the SwarmClaw control plane, so your agent/task/session actions apply across the fleet view. Use it when you need status and task coordination across more than one OpenClaw instance.

Commands

Agents:

swarmclaw agents list — list all agents with IDs, names, providers, status
```
swarmclaw agents list --raw
```
swarmclaw agents get <agentId> — get full details for a specific agent
```
swarmclaw agents get <agentId> --raw
```

Tasks:

swarmclaw tasks create --title "..." --description "..." --agent-id <agentId> — create and assign a task

swarmclaw tasks create --title "Analyse competitor pricing" --description "Research and compare competitor pricing strategies, identify gaps and opportunities" --agent-id <agentId> --raw

swarmclaw tasks list — list all tasks with status
```
swarmclaw tasks list --raw
```

Sessions:

swarmclaw sessions list — list active sessions
```
swarmclaw sessions list --raw
```
swarmclaw sessions chat <id> --message "..." — send a message to an existing session
```
swarmclaw sessions chat <id> --message "Give me a status update" --raw
```

Memory:

swarmclaw memory maintenance — run memory maintenance analysis
```
swarmclaw memory maintenance --raw
```

Diagnostics:

swarmclaw setup doctor — run system diagnostics and report issues
```
swarmclaw setup doctor --raw
```

Version:

swarmclaw version check — check current version and update status
```
swarmclaw version check --raw
```

Rules

Always verify an agent ID exists by running agents list before creating tasks for it.
Never pass the access key as a CLI argument in visible output — rely on the SWARMCLAW_ACCESS_KEY env var.
If the SwarmClaw instance is unreachable, run swarmclaw setup doctor and report the findings before retrying.
When the user says "dispatch work", "get X agent to do Y", or "assign this to", use tasks create to assign the work to the appropriate agent.
When the user asks "what's running", "agent status", or "what are my agents doing", combine output from agents list and sessions list to give a full picture.
For complex multi-step orchestration, create individual tasks rather than chaining commands.
Use --raw/--json output modes when you need to parse command output for further processing.
Do not run commands that modify or delete agents without explicit user confirmation.

Examples

Dispatch work to another agent: User says: "Get my research agent to analyse competitor pricing"

swarmclaw agents list --raw
swarmclaw tasks create --title "Analyse competitor pricing" --description "Research and compare competitor pricing strategies, identify gaps and opportunities" --agent-id <research-agent-id> --raw

Then confirm the task was created and the agent will pick it up.

Check fleet status: User says: "What are all my agents doing right now?"

swarmclaw agents list --raw
swarmclaw sessions list --raw

Then summarize which agents are idle, which have active sessions, and any tasks in progress.

Run diagnostics: User says: "Something seems wrong with SwarmClaw"

swarmclaw setup doctor --raw

Then report issues found and suggest fixes based on the doctor output.

Check agents across gateways: User says: "What's happening across all my OpenClaw instances?"

swarmclaw agents list --raw

Then use the output to identify agents across multiple gateways with their connection status, and summarize which gateways are healthy and which agents are active on each.

🇯🇵 日本人クリエイター向け解説

🎯 このSkillでできること

📦 インストール方法 (3ステップ)