From agent-patterns
Decompose a task into an orchestrator and worker architecture. Use this skill when asked to "plan an agent", "design a multi-agent system", "break down this task for agents", or any request that involves delegating work across multiple Claude instances or tool-calling pipelines.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-patterns:agent-planThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Decompose the given task into an **orchestrator + worker** multi-agent architecture.
Decompose the given task into an orchestrator + worker multi-agent architecture.
Read the task description and identify:
The orchestrator is responsible for:
Define the orchestrator's interface:
Orchestrator Input: <describe the input schema>
Orchestrator Output: <describe the output schema>
State managed: <list what the orchestrator tracks>
For each subtask, define a worker with a narrow scope:
Worker: <name>
Input: <what it receives from the orchestrator>
Output: <what it returns to the orchestrator>
Scope: <one sentence describing exactly what it does>
Can run in parallel with: <list other workers or "none">
Principles for good workers:
Produce a text-based sequence diagram showing the message flow:
User -> Orchestrator: <task>
Orchestrator -> Worker A: <subtask 1>
Worker A -> Orchestrator: <result 1>
Orchestrator -> Worker B: <subtask 2> (can run after step 1, or in parallel)
Worker B -> Orchestrator: <result 2>
Orchestrator -> User: <final output>
For each worker, define what the orchestrator should do if the worker fails:
Summarize the architecture as a concise table:
| Component | Role | Input | Output | Parallelizable |
|---|---|---|---|---|
| Orchestrator | ... | ... | ... | N/A |
| Worker A | ... | ... | ... | Yes/No |
| Worker B | ... | ... | ... | Yes/No |
Ask the user to confirm the decomposition before implementation begins.
npx claudepluginhub ats-kinoshita-iso/agent-workshop --plugin agent-patternsDesigns and implements multi-agent LLM systems using orchestrator patterns, parallel coordination, pipelines, hierarchical delegation, communication, and failure handling. For agent workflows and debugging failures.
Analyzes user tasks to recommend and execute optimal agent orchestration patterns: Sequential Pipeline, Parallel Subagent, Team Mode, Ralph Loop. For complex multi-step tasks or /agent-orchestrate invocation.
Patterns for multi-agent coordination, task decomposition, agent handoffs, and orchestration topology selection. Use when splitting large tasks across sub-agents or debugging agent systems.