메뉴
Coordinate multiple AI agents for complex tasks — decomposition, delegation, and synthesis
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/fabioc-aloha/ChessCoach --skill multi-agent-orchestration명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
End-to-end academic paper drafting for CHI, HBR, journals, and conferences with venue-specific templates, drafting workflows, and revision strategies.
Detect, create, and manage the AI-Memory fleet communication channel. Fires on bootstrap, session start (announcements), and feedback writes.
Generate on-brand Alex — ACT Edition SVG banners for documents (READMEs, plans, notes, release artifacts)
SOC 직업 분류 기준
SKILL.md 표시 중
| type | skill |
| lifecycle | stable |
| inheritance | inheritable |
| name | multi-agent-orchestration |
| description | Coordinate multiple AI agents for complex tasks — decomposition, delegation, and synthesis |
| tier | standard |
| user-invokable | false |
| applyTo | **/*agent*,**/*orchestrat*,**/*multi*,**/*workflow*,**/*subagent* |
| currency | 2026-04-22T00:00:00.000Z |
Decompose complex problems into agent-appropriate subtasks, delegate effectively, and synthesize results.
Multi-agent patterns are evolving rapidly. This skill captures stable patterns while acknowledging the field is in flux.
Refresh triggers:
Last validated: February 2026
| Scenario | Single Agent | Multi-Agent |
|---|---|---|
| Simple code edit | ✅ | ❌ Overkill |
| Multi-file refactor | ✅ (if capable model) | ⚠️ Consider |
| Research + implement | ⚠️ Long context | ✅ Decompose |
| Cross-domain task | ❌ Context overload | ✅ Specialists |
| Parallel independent work | ❌ Sequential | ✅ Parallel agents |
| Role | Responsibility | Example |
|---|---|---|
| Orchestrator | Decompose, delegate, synthesize | Main chat session |
| Specialist | Deep expertise in one domain | Security reviewer agent |
| Worker | Execute well-defined subtask | "Find all usages of X" |
| Critic | Validate, review, improve | Code review agent |
Split task into independent subtasks that can run simultaneously.
┌─────────────────┐
│ Orchestrator │
└───────┬─────────┘
│ decompose
┌────┴────┬────────┐
▼ ▼ ▼
┌─────┐ ┌─────┐ ┌─────┐
│ A1 │ │ A2 │ │ A3 │ (parallel)
└──┬──┘ └──┬──┘ └──┬──┘
└────────┼────────┘
▼
synthesize
When to use:
Example: "Search for security issues in auth, api, and database modules"
Chain agents where each builds on previous output.
┌─────────────────┐
│ Orchestrator │
└───────┬─────────┘
▼
┌───────┐
│ A1 │ → research
└───┬───┘
▼
┌───────┐
│ A2 │ → analyze
└───┬───┘
▼
┌───────┐
│ A3 │ → implement
└───────┘
When to use:
Example: "Research best practices → Design API → Implement → Review"
Orchestrator delegates to sub-orchestrators who manage workers.
┌─────────────────┐
│ Root Orch │
└───────┬─────────┘
┌────┴────┐
▼ ▼
┌─────┐ ┌─────┐
│SubO1│ │SubO2│ (sub-orchestrators)
└──┬──┘ └──┬──┘
┌─┴─┐ ┌──┴──┐
▼ ▼ ▼ ▼
┌─┐ ┌─┐ ┌─┐ ┌─┐
│W│ │W│ │W│ │W│ (workers)
└─┘ └─┘ └─┘ └─┘
When to use:
When delegating to a subagent, specify:
| Element | Purpose | Example |
|---|---|---|
| Context | What they need to know | "Working on Alex VS Code extension" |
| Scope | Clear boundaries | "Only look in /src/services" |
| Output | Expected format | "Return JSON with findings" |
| Constraints | What NOT to do | "Don't modify files, only report" |
**Task:** [One-sentence objective]
**Context:**
- Project: [name/type]
- Relevant files: [list]
- What's already done: [state]
**Scope:**
- DO: [specific actions]
- DON'T: [boundaries]
**Expected Output:**
[Format and content expectations]
**Success Criteria:**
[How you'll know it's done right]
| Pattern | When | How |
|---|---|---|
| Concatenate | Independent results | Simple append |
| Deduplicate | Overlapping searches | Hash/compare |
| Vote | Multiple opinions | Majority wins |
| Synthesize | Diverse perspectives | LLM summary |
| Validate | Critical decisions | Critic agent reviews |
When agents disagree:
Identify conflict type
Resolution strategies
runSubagent EffectivelyThe runSubagent tool enables orchestration within VS Code:
// Good: Clear task with expected output
await runSubagent({
prompt: `Search the codebase for all error handling patterns.
Return a JSON array of: {file, line, pattern, quality}`,
description: "Find error patterns"
user-invokable: false
});
// Bad: Vague delegation
await runSubagent({
prompt: "Look for problems in the code", // Too vague
description: "Find issues"
user-invokable: false
});
| Scenario | Approach |
|---|---|
| Simple search | Direct grep_search |
| Complex multi-step search | runSubagent |
| Single file edit | Direct replace_string_in_file |
| Multi-file coordinated change | Consider subagent for planning |
| Research + implementation | Subagent for research, direct for implementation |
Problem: Using multiple agents for simple tasks Symptom: Slower, more expensive, no quality gain Fix: Trust capable models for multi-step tasks up to complexity threshold
Problem: Agents lack needed information Symptom: Repeated clarification requests, wrong assumptions Fix: Front-load context in delegation prompt
Problem: Raw agent outputs dumped on user Symptom: User must manually integrate results Fix: Plan synthesis before decomposition
Problem: Agent A needs B's output, B needs A's output Symptom: Deadlock or infinite loops Fix: Identify and break cycles in task graph
| Framework | Strength | Use Case |
|---|---|---|
| LangGraph | State machines, cycles | Complex workflows |
| AutoGen | Conversation patterns | Research, debate |
| CrewAI | Role-based teams | Business processes |
| VS Code Agents | IDE integration | Code tasks |
| Semantic Kernel | .NET native | Enterprise C# |
Master Alex can coordinate heirs for cross-platform tasks:
Master Alex (orchestrator)
├── VS Code Heir → code analysis
├── M365 Heir → document synthesis
└── Global Knowledge → pattern matching
When Alex runs Skill Selection Optimization (SSO), it's a form of self-orchestration:
When designing multi-agent workflows:
Multi-agent orchestration is powerful but not always necessary. Start simple, add agents when complexity demands it.