swarmclaw
정보
이 Claude Skill은 SwarmClaw 멀티 에이전트 오케스트레이션 플랫폼을 관리하기 위한 CLI 접근을 제공하며, 에이전트, 작업, 일정, 메모리 및 40개 이상의 다른 명령어 그룹을 제어할 수 있습니다. 작업 발송, 에이전트 조정, 진단 실행, 대시보드 인스턴스 간 오케스트레이션에 사용됩니다. CLI는 SwarmClaw REST API에 직접 매핑되며 `SWARMCLAW_ACCESS_KEY` 환경 변수가 필요합니다.
빠른 설치
Claude Code
추천npx skills add swarmclawai/swarmclaw -a claude-code/plugin add https://github.com/swarmclawai/swarmclawgit clone https://github.com/swarmclawai/swarmclaw.git ~/.claude/skills/swarmclawClaude Code에서 이 명령을 복사하여 붙여넣어 스킬을 설치하세요
문서
Overview
SwarmClaw is a self-hosted AI agent orchestration platform. This skill gives you CLI access to manage agents, tasks, chats, chatrooms, goals, schedules, memory, wallets, connectors, autonomy controls, webhooks, extensions, and more across one or many SwarmClaw instances.
The CLI maps 1:1 to the SwarmClaw REST API. Every command follows the pattern:
swarmclaw <group> <action> [id] [--flags]
Install the CLI:
npm i -g @swarmclawai/swarmclaw
Setup
- Install:
npm i -g @swarmclawai/swarmclaw - Set env var:
export SWARMCLAW_ACCESS_KEY=<your-key>(shown in terminal on first run) - Default URL:
http://localhost:3456— override withSWARMCLAW_URLor--base-url - Full docs: https://swarmclaw.ai/docs
Output Modes
Use machine-readable output when parsing results:
--json— structured JSON output (preferred for all API-mapped commands)--raw— plain text output (legacy commands)
Filter results with --query key=value:
swarmclaw tasks list --json --query status=in_progress
swarmclaw memory list --json --query agentId=abc123 --query q=pricing
swarmclaw runs list --json --query sessionId=xyz --query limit=10
Core Commands
Agents
Manage your agent fleet — create, configure, monitor, and clone agents.
# List all agents
swarmclaw agents list --json
# Get full agent details
swarmclaw agents get <agentId> --json
# Create an agent (pass JSON body via --data)
swarmclaw agents create --data '{"name":"Research Bot","provider":"anthropic","model":"claude-sonnet-4-20250514"}' --json
# Update agent config
swarmclaw agents update <agentId> --data '{"name":"Updated Name","soul":"You are a research assistant"}' --json
# Get live agent status (running chats, current task, etc.)
swarmclaw agents status <agentId> --json
# Clone an agent with all its config
swarmclaw agents clone <agentId> --json
# Bulk update multiple agents
swarmclaw agents bulk-update --data '{"ids":["id1","id2"],"changes":{"provider":"openai"}}' --json
Other: delete, trash, restore, purge, thread
Tasks
Create, assign, and track work items on the task board.
# List all tasks (filterable by status, agent, project)
swarmclaw tasks list --json
swarmclaw tasks list --json --query status=in_progress --query agentId=<id>
# Create and assign a task
swarmclaw tasks create --data '{"title":"Analyse competitor pricing","description":"Research and compare competitor pricing strategies","agentId":"<agentId>"}' --json
# Get task details
swarmclaw tasks get <taskId> --json
# Update task status or fields
swarmclaw tasks update <taskId> --data '{"status":"completed"}' --json
# Get task board metrics (24h, 7d, or 30d)
swarmclaw tasks metrics --json --query range=7d
# Import GitHub issues as tasks
swarmclaw tasks import-github --data '{"repo":"owner/repo","labels":["bug"],"agentId":"<id>"}' --json
# Bulk update tasks
swarmclaw tasks bulk --data '{"ids":["id1","id2"],"changes":{"status":"todo"}}' --json
Other: delete, approve, claim, purge
Chats
Manage agent chat sessions — create, send messages, stream responses, and control execution.
Terminology:
chatsin the CLI = "sessions" in the SwarmClaw UI. Both refer to the same thing.
# List chats
swarmclaw chats list --json
# Create a new chat for an agent
swarmclaw chats create --data '{"agentId":"<agentId>"}' --json
# Send a message and stream the agent response (SSE)
swarmclaw chats chat <chatId> --data '{"message":"Give me a status update"}' --json
# Get chat message history
swarmclaw chats messages <chatId> --json
# Stop a running chat
swarmclaw chats stop <chatId> --json
# Clear chat history
swarmclaw chats clear <chatId> --json
# Append a user message without triggering a response
swarmclaw chats messages-send <chatId> --data '{"role":"user","content":"Note: budget is $5000"}' --json
Other: get, update, delete, delete-many, edit-resend, retry, queue, queue-add, queue-clear, heartbeat-disable-all, deploy, devserver, checkpoints
Chatrooms
Multi-agent collaborative spaces where multiple agents discuss and work together.
# List chatrooms
swarmclaw chatrooms list --json
# Create a chatroom
swarmclaw chatrooms create --data '{"name":"Strategy Room","description":"Multi-agent planning"}' --json
# Add agents to a chatroom
swarmclaw chatrooms add-member <chatroomId> --data '{"agentId":"<agent1>"}' --json
swarmclaw chatrooms add-member <chatroomId> --data '{"agentId":"<agent2>"}' --json
# Send a message — all member agents respond
swarmclaw chatrooms chat <chatroomId> --data '{"message":"What should our Q2 strategy be?"}' --json
# Remove an agent from the chatroom
swarmclaw chatrooms remove-member <chatroomId> --data '{"agentId":"<agentId>"}' --json
Other: get, update, delete, react, pin, moderate
Schedules
Automate recurring agent work on a cron schedule.
# List schedules
swarmclaw schedules list --json
# Create a schedule (cron syntax)
swarmclaw schedules create --data '{"name":"Daily Report","agentId":"<id>","cron":"0 9 * * *","message":"Generate the daily summary report"}' --json
# Update schedule
swarmclaw schedules update <scheduleId> --data '{"enabled":false}' --json
# Trigger a schedule immediately
swarmclaw schedules run <scheduleId> --json
# Delete a schedule
swarmclaw schedules delete <scheduleId> --json
Memory
Store and retrieve long-term agent memories for context persistence.
# List memories (searchable)
swarmclaw memory list --json
swarmclaw memory list --json --query q=pricing --query agentId=<id>
# Create a memory entry
swarmclaw memory create --data '{"agentId":"<id>","content":"Client prefers formal tone","type":"preference"}' --json
# Update a memory
swarmclaw memory update <memoryId> --data '{"content":"Updated content"}' --json
# Delete a memory
swarmclaw memory delete <memoryId> --json
# Analyse memory for dedup/prune candidates
swarmclaw memory maintenance --json
# Run the maintenance (dedupe/prune)
swarmclaw memory maintenance-run --json
Other: get, graph
Goals
Hierarchical goal management — organization, team, project, agent, and task-level goals with parent-child chains.
# List all goals
swarmclaw goals list --json
# Create a goal
swarmclaw goals create --data '{"title":"Increase revenue 20%","level":"organization","description":"Q2 revenue target"}' --json
# Get goal details
swarmclaw goals get <goalId> --json
# Update a goal
swarmclaw goals update <goalId> --data '{"status":"in_progress"}' --json
# Delete a goal
swarmclaw goals delete <goalId> --json
Knowledge
Manage knowledge sources — manual text, files, and URLs that ground agent responses with citations.
# List all knowledge entries
swarmclaw knowledge list --json
# Search knowledge
swarmclaw knowledge list --json --query q=pricing
# Create a manual knowledge entry
swarmclaw knowledge create --data '{"title":"Pricing Policy","content":"Standard pricing is...","kind":"manual","tags":["pricing"]}' --json
# Upload a file as knowledge
swarmclaw knowledge upload ./docs/handbook.pdf --json
# Get full source detail (metadata + indexed chunks)
swarmclaw knowledge source-get <sourceId> --json
# List knowledge sources with summaries
swarmclaw knowledge sources --json
# Re-sync a file/URL source
swarmclaw knowledge source-sync <sourceId> --json
# Archive a source
swarmclaw knowledge source-archive <sourceId> --data '{"reason":"outdated"}' --json
# Restore an archived source
swarmclaw knowledge source-restore <sourceId> --json
# Mark a source as superseded by another
swarmclaw knowledge source-supersede <sourceId> --data '{"supersededBySourceId":"<newSourceId>"}' --json
# Get knowledge hygiene summary (stale, duplicate, broken sources)
swarmclaw knowledge hygiene --json
# Run hygiene maintenance (auto-sync stale, archive duplicates)
swarmclaw knowledge hygiene-run --json
Other: get, update, delete, source-create, source-update, source-delete
Search
Global cross-resource search across agents, tasks, chats, schedules, webhooks, and skills.
swarmclaw search query --json --query q=pricing
System
System health, version, and usage information.
# Health check (lightweight, safe for polling)
swarmclaw system status --json
# Check current version and updates
swarmclaw system version --json
# View resource usage
swarmclaw system usage --json
More Commands
These groups are available but used less frequently by agents. Use swarmclaw <group> --help for full details.
| Group | Description | Key commands |
|---|---|---|
| connectors | Chat connectors (Discord, Slack, Telegram, etc.) | list, get, create, update, delete, start, stop, repair, health, doctor |
| autonomy | Supervisor incidents, reflections, emergency stop | incidents, reflections, estop, estop-set, guardian-restore |
| approvals | Human-in-the-loop approval gates | list, resolve |
| webhooks | Inbound webhook triggers and delivery history | list, get, create, update, delete, trigger, history |
| wallets | Agent wallets and transaction management | list, get, create, update, delete, send, approve, transactions, balance-history |
| providers | LLM provider configs and model overrides | list, get, create, update, delete, configs, ollama, models, models-set |
| gateways | OpenClaw gateway profiles and health checks | list, create, update, delete, health |
| credentials | Encrypted provider credentials | list, get, create, delete |
| secrets | Encrypted secret vault | list, get, create, update, delete |
| notifications | In-app notification center | list, create, clear, mark-read, delete |
| extensions | Extension marketplace and config | list, set, install, marketplace, settings-get, settings-set, builtins |
| knowledge | Knowledge source management and hygiene | list, get, create, update, delete, upload, sources, source-get, source-create, source-update, source-delete, source-archive, source-restore, source-supersede, source-sync, hygiene, hygiene-run |
| skills | Reusable skill management | list, get, create, update, delete, import |
| learned-skills | Agent-scoped learned skill review | list, promote, dismiss, delete, review-counts |
| skill-suggestions | AI-generated skill recommendations | list, draft, approve, reject |
| external-agents | External agent registration and heartbeat | list, create, update, delete, heartbeat |
| delegation-jobs | Cross-agent delegation job tracking | list |
| portability | Config import/export between installs | export, import |
| settings | App-level configuration | get, update |
| runs | Chat run queue and execution history | list, get, events |
| activity | Activity feed events | list (supports --query entityType=, --query action=) |
| daemon | Background daemon lifecycle | status, start, stop, health-check |
| logs | Application logs | list (supports --query lines=200, --query level=INFO), clear |
| setup | Diagnostics and provider validation | doctor, check-provider |
| documents | Document management and revisions | list, get, create, update, delete, revisions |
| uploads | Uploaded artifact management | list, get, delete, delete-many |
| clawhub | Browse and install ClawHub skills | search, preview, install |
| openclaw | OpenClaw gateway control, deploy, sync (30+ subcommands) | discover, gateway-status, deploy-*, remote-*, skills, sync |
Rules
- Always verify an agent ID exists by running
agents listbefore creating tasks for it. - Never pass the access key as a CLI argument in visible output — rely on the
SWARMCLAW_ACCESS_KEYenv var. - If the SwarmClaw instance is unreachable, run
swarmclaw setup doctor --jsonand report the findings before retrying. - When the user says "dispatch work", "get X agent to do Y", or "assign this to", use
tasks createto assign the work. - When the user asks "what's running", "agent status", or "what are my agents doing", combine output from
agents listandchats listto give a full picture. - For complex multi-step orchestration, create individual tasks rather than chaining commands.
- Prefer
--jsonoutput mode for all commands. Use--rawonly for legacy commands that don't support--json. - Do not run commands that modify or delete agents without explicit user confirmation.
- Use
system statusfor quick health checks — it's lightweight and safe for repeated polling. - Use
search query --query q=<term>to discover resources across agents, tasks, chats, schedules, and skills. - The CLI group
chats= "sessions" in the SwarmClaw UI. The legacysessionsalias is no longer available; always usechats. - Use
autonomy estop-setto engage emergency stop across all autonomous agents when safety action is needed.
Examples
Dispatch work to another agent
User says: "Get my research agent to analyse competitor pricing"
swarmclaw agents list --json
# Find the research agent's ID from the output
swarmclaw tasks create --data '{"title":"Analyse competitor pricing","description":"Research and compare competitor pricing strategies, identify gaps and opportunities","agentId":"<research-agent-id>"}' --json
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 --json
swarmclaw chats list --json
swarmclaw tasks list --json --query status=in_progress
Then summarize which agents are idle, which have active chats, and any tasks in progress.
Run diagnostics
User says: "Something seems wrong with SwarmClaw"
swarmclaw system status --json
swarmclaw setup doctor --json
Check the health summary first (fast), then run full diagnostics if issues are found.
Multi-agent chatroom collaboration
User says: "Get my strategy and research agents to brainstorm Q2 plans together"
swarmclaw agents list --json
# Identify the strategy and research agent IDs
swarmclaw chatrooms create --data '{"name":"Q2 Strategy Brainstorm"}' --json
swarmclaw chatrooms add-member <chatroomId> --data '{"agentId":"<strategy-agent-id>"}' --json
swarmclaw chatrooms add-member <chatroomId> --data '{"agentId":"<research-agent-id>"}' --json
swarmclaw chatrooms chat <chatroomId> --data '{"message":"Brainstorm our Q2 growth strategy. Consider market trends, competitor moves, and our current strengths."}' --json
Schedule recurring work
User says: "Have my reporting agent send a daily summary every morning at 9am"
swarmclaw agents list --json
# Find the reporting agent's ID
swarmclaw schedules create --data '{"name":"Daily Morning Summary","agentId":"<reporting-agent-id>","cron":"0 9 * * *","message":"Generate and send the daily summary report covering key metrics, alerts, and action items from the last 24 hours"}' --json
Emergency stop all autonomous agents
User says: "Stop everything now!"
swarmclaw autonomy estop-set --data '{"engaged":true}' --json
Then confirm the emergency stop is engaged and all autonomous execution has halted.
Set up a goal hierarchy
User says: "Create a Q2 revenue goal and link my sales agent's tasks to it"
swarmclaw goals create --data '{"title":"Increase Q2 revenue 20%","level":"organization","description":"Hit $1.2M ARR by end of Q2"}' --json
# Use the returned goalId to link agent tasks
swarmclaw tasks list --json --query agentId=<sales-agent-id>
swarmclaw tasks update <taskId> --data '{"goalId":"<goalId>"}' --json
Discovery
This skill covers the most-used commands. The full CLI has 50+ command groups. To explore:
swarmclaw --help # List all command groups
swarmclaw <group> --help # List commands in a group
swarmclaw <group> <action> --help # Show flags for a specific command
GitHub 저장소
연관 스킬
llamaguard
기타LlamaGuard는 폭력 및 혐오 발언 등 6가지 안전 범주에서 LLM 입력과 출력을 조정하기 위한 Meta의 70-80억 파라미터 모델입니다. 94-95% 정확도를 제공하며 vLLM, Hugging Face 또는 Amazon SageMaker를 사용해 배포할 수 있습니다. 이 기술을 사용하여 AI 애플리케이션에 콘텐츠 필터링 및 안전 가드레일을 손쉽게 통합하세요.
cost-optimization
기타이 Claude Skill은 리소스 적정화, 태깅 전략, 지출 분석을 통해 개발자들이 클라우드 비용을 최적화할 수 있도록 지원합니다. AWS, Azure, GCP에서 클라우드 비용을 절감하고 비용 거버넌스를 구현하기 위한 프레임워크를 제공합니다. 인프라 비용을 분석하거나, 리소스를 적정화하거나, 예산 제약을 충족해야 할 때 사용하세요.
quantizing-models-bitsandbytes
기타이 스킬은 bitsandbytes를 사용하여 LLM을 8비트 또는 4비트 정밀도로 양자화하며, 최소한의 정확도 손실로 50-75%의 메모리 감소를 달성합니다. 제한된 GPU 메모리에서 더 큰 모델을 실행하거나 추론을 가속화하는 데 이상적이며, INT8, NF4, FP4와 같은 형식을 지원합니다. 이 스킬은 HuggingFace Transformers와 통합되어 QLoRA 학습 및 8비트 옵티마이저를 가능하게 합니다.
dispatching-parallel-agents
기타이 Claude Skill은 3개 이상의 독립적인 문제를 동시에 조사하고 해결하기 위해 다중 에이전트를 배치합니다. 공유 상태나 의존성 없이 해결 가능한 무관련 장애 시나리오에 맞게 설계되었습니다. 핵심 기능은 병렬 문제 해결로, 각 독립 문제 영역마다 하나의 에이전트를 할당하여 효율성을 극대화합니다.
