{"skill":{"slug":"agent-orchestrator-molter","displayName":"Agent Orchestrator","summary":"Multi-agent orchestration with 5 proven patterns - Work Crew, Supervisor, Pipeline, Council, and Auto-Routing","description":"---\nname: agent-orchestrator\nversion: 1.0.5\nauthor: molter-white\ndescription: Multi-agent orchestration with 5 proven patterns - Work Crew, Supervisor, Pipeline, Council, and Auto-Routing\nlicense: MIT\ntags: multi-agent,orchestration,automation,productivity,ai-workflow\ncompatibility: OpenClaw 0.8+\n---\n\n# agent-orchestrator\n\nMulti-agent orchestration for OpenClaw. Implements 5 proven patterns for coordinating multiple AI agents: Work Crew, Supervisor, Pipeline, Expert Council, and Auto-Routing.\n\n**USE WHEN:**\n- A task can be parallelized for speed or redundancy (Work Crew)\n- Complex tasks need dynamic planning and delegation (Supervisor)\n- Work follows a predictable sequence of stages (Pipeline)\n- Cross-domain input is needed from multiple specialists (Expert Council)\n- Mixed task types need automatic routing to appropriate specialists (Auto-Routing)\n- Research tasks require breadth-first exploration of multiple angles\n- High-stakes decisions need confidence through multiple perspectives\n\n**DON'T USE WHEN:**\n- Simple tasks that fit in one agent's context window (use main session instead)\n- Sequential tasks with no parallelization opportunity (use regular tool calls)\n- One-shot deterministic tasks (use single agent)\n- Tasks requiring real-time inter-agent conversation (this uses async spawning)\n- Tasks where 15x token cost cannot be justified\n- Quick/simple tasks where coordination overhead exceeds benefit\n\n**Outputs:**\n- Aggregated results from multiple parallel agents\n- Synthesized consensus recommendations\n- Routing decisions to appropriate specialists\n- Structured output from staged processing\n\n## Decision Matrix\n\n| Pattern | Use When | Avoid When |\n|---------|----------|------------|\n| **crew** | Same task from multiple angles, verification, research breadth | Results cannot be easily compared/merged |\n| **supervise** | Dynamic decomposition needed, complex planning | Fixed workflow, simple delegation |\n| **pipeline** | Well-defined sequential stages, content creation | Path needs runtime adaptation |\n| **council** | Cross-domain expertise, risk assessment, policy review | Single-domain task, need fast consensus |\n| **route** | Mixed workload types, automatic classification | Task type is already known |\n\n## Auto-Routing Pattern\n\nThe route command analyzes tasks and automatically classifies them by type, then routes to the appropriate specialist:\n\n```bash\n# Basic routing\nclaw agent-orchestrator route --task \"Write Python parser\"\n\n# With custom specialist pool\nclaw agent-orchestrator route \\\n  --task \"Analyze data and create report\" \\\n  --specialists \"analyst,data,writer\"\n\n# Force specific specialist\nclaw agent-orchestrator route \\\n  --task \"Something complex\" \\\n  --force coder\n```\n\n### Confidence Thresholds\n\n- **High confidence (>0.85)**: Auto-route immediately\n- **Good confidence (0.7-0.85)**: Propose with confirmation option\n- **Moderate confidence (0.5-0.7)**: Show top alternatives\n- **Low confidence (<0.5)**: Request clarification\n\nAvailable specialists: coder, researcher, writer, analyst, planner, reviewer, creative, data, devops, support\n\n## Common Workflows\n\n```bash\n# Parallel research with consensus\nclaw agent-orchestrator crew \\\n  --task \"Research Bitcoin Lightning 2026 adoption\" \\\n  --agents 4 \\\n  --perspectives technical,business,security,competitors \\\n  --converge consensus\n\n# Best-of redundancy for critical analysis\nclaw agent-orchestrator crew \\\n  --task \"Audit this smart contract for vulnerabilities\" \\\n  --agents 3 \\\n  --converge best-of\n\n# Supervisor-managed code review\nclaw agent-orchestrator supervise \\\n  --task \"Refactor authentication module\" \\\n  --workers coder,reviewer,tester \\\n  --strategy adaptive\n\n# Staged content pipeline\nclaw agent-orchestrator pipeline \\\n  --stages research,draft,review,finalize \\\n  --input \"topic: AI agent adoption trends\"\n\n# Expert council for decision\nclaw agent-orchestrator council \\\n  --question \"Should we publish this blog post about unreleased features?\" \\\n  --experts skeptic,ethicist,strategist \\\n  --converge consensus \\\n  --rounds 2\n\n# Auto-route mixed tasks\nclaw agent-orchestrator route \\\n  --task \"Write Python function to analyze CSV data\" \\\n  --specialists coder,researcher,writer,analyst\n\n# Force route to specific specialist\nclaw agent-orchestrator route \\\n  --task \"Debug authentication error\" \\\n  --force coder \\\n  --confidence-threshold 0.9\n\n# Route and output as JSON for scripting\nclaw agent-orchestrator route \\\n  --task $TASK \\\n  --format json \\\n  --specialists \"coder,data,analyst\"\n```\n\n## Negative Examples\n\n**DON'T: Use crew for simple single-answer questions**\n```bash\n# WRONG: Wasteful for simple facts\nclaw agent-orchestrator crew --task \"What is 2+2?\" --agents 3\n\n# RIGHT: Use main session directly\nWhat is 2+2?\n```\n\n**DON'T: Use supervise when pipeline suffices**\n```bash\n# WRONG: Over-engineering fixed workflows\nclaw agent-orchestrator supervise --task \"Draft, edit, publish\"\n\n# RIGHT: Use pipeline for fixed sequences\nclaw agent-orchestrator pipeline --stages draft,edit,publish\n```\n\n**DON'T: Route when task type is obvious**\n```bash\n# WRONG: Unnecessary classification overhead\nclaw agent-orchestrator route --task \"Write Python code\"\n\n# RIGHT: Direct to appropriate specialist\nclaw agent-orchestrator crew --pattern code --task \"Write Python code\"\n```\n\n**DON'T: Use multi-agent for very small context tasks**\n```bash\n# WRONG: Coordination overhead exceeds value\nclaw agent-orchestrator crew --task \"Fix typo\" --agents 2\n\n# RIGHT: Single agent or direct edit\nedit file.py \"typo\" \"correct\"\n```\n\n## Token Cost Warning\n\nMulti-agent patterns use approximately 15x more tokens than single-agent interactions. Use only for high-value tasks where quality improvement justifies the cost. See Anthropic research: token usage explains 80% of performance variance in complex tasks.\n\n## Dependencies\n\n- Python 3.8+\n- OpenClaw sessions_spawn capability\n- OpenClaw sessions_list capability\n- OpenClaw sessions_history capability\n\n## Files\n\n- `__main__.py` - CLI entry point\n- `crew.py` - Work Crew pattern implementation\n- `supervise.py` - Supervisor pattern (Phase 2)\n- `council.py` - Expert Council pattern (Phase 2)\n- `pipeline.py` - Pipeline pattern (Phase 2)\n- `route.py` - Auto-Routing pattern (Phase 2)\n- `utils.py` - Shared utilities for session management\n\n## Status\n\n- MVP: Work Crew pattern implemented\n- **Phase 2: 100% Complete**\n  - [x] Supervisor pattern implemented - dynamic task decomposition and worker delegation\n  - [x] Pipeline pattern implemented - sequential staged processing with validation gates\n  - [x] Council pattern implemented - multi-expert deliberation with convergence methods\n  - [x] Route pattern implemented - intelligent task classification and specialist routing\n\n## References\n\n- Anthropic Multi-Agent Research System\n- LangGraph Supervisor Pattern\n- CrewAI Framework\n- AutoGen Conversational Agents\n","tags":{"ai-workflow":"1.0.5","automation":"1.0.5","latest":"1.0.5","multi-agent":"1.0.5","orchestration":"1.0.5","productivity":"1.0.5"},"stats":{"comments":0,"downloads":1256,"installsAllTime":47,"installsCurrent":1,"stars":0,"versions":5},"createdAt":1771196681574,"updatedAt":1778491549296},"latestVersion":{"version":"1.0.5","createdAt":1771199517743,"changelog":"Scanner false-positive reduction: reworded safety signatures/preamble without lowering protections.","license":null},"metadata":null,"owner":{"handle":"variable190","userId":"s1764rj21v083gp2g7j00qrsys885576","displayName":"Dan","image":"https://avatars.githubusercontent.com/u/50242220?v=4"},"moderation":null}