LeoYeAI/agent-master-skills / skills/agent-avengers
agent-avengers
Verified올인원 멀티에이전트 오케스트레이션. 복잡한 태스크를 자동으로 분해하고, 전문 에이전트들을 즉석에서 생성/스폰하여 병렬 처리 후 결과를 통합합니다. TRIGGERS: avengers assemble, 어벤저스, agent-avengers, 멀티에이전트 자동화, 에이전트 팀 구성, 자동 에이전트
| File | Kind | Size |
|---|---|---|
| ANNOUNCEMENT.md | markdown | 1.7 KB |
| README-kr.md | markdown | 1.0 KB |
| README.md | markdown | 2.7 KB |
| SKILL.md | skill md | 22.9 KB |
| _meta.json | config | 283 B |
| examples/app-development.json | config | 1.3 KB |
| examples/competitor-analysis.json | config | 1.1 KB |
| package.json | manifest | 653 B |
| pytest.ini | config | 125 B |
| scripts/assemble.py | script | 10.0 KB |
| scripts/consolidate.py | script | 6.4 KB |
| scripts/execute.py | script | 6.5 KB |
| scripts/monitor.py | script | 6.3 KB |
| tests/__init__.py | script | 23 B |
| tests/test_assemble.py | script | 7.8 KB |
| tests/test_consolidate.py | script | 6.4 KB |
scripts/assemble.py
View on GitHub#!/usr/bin/env python3
"""
Agent Avengers - Assemble Script
태스크 분해 → 에이전트 배정 → 스폰/디스패치
"""
import json
import os
import sys
import argparse
from datetime import datetime
from pathlib import Path
# 워크스페이스 경로
WORKSPACE = os.environ.get("AVENGERS_WORKSPACE", os.path.expanduser("~/.openclaw/workspace"))
MISSION_DIR = Path(WORKSPACE) / "avengers-missions"
# 에이전트 타입 정의
AGENT_TYPES = {
"researcher": {
"emoji": "🔬",
"model": "sonnet",
"timeout": 1800,
"keywords": ["조사", "리서치", "검색", "수집", "분석"]
},
"analyst": {
"emoji": "🔍",
"model": "opus",
"timeout": 1200,
"keywords": ["분석", "패턴", "인사이트", "평가"]
},
"writer": {
"emoji": "🖊️",
"model": "sonnet",
"timeout": 900,
"keywords": ["작성", "문서", "리포트", "콘텐츠", "글"]
},
"coder": {
"emoji": "💻",
"model": "opus",
"timeout": 2400,
"keywords": ["코드", "개발", "구현", "API", "프로그래밍"]
},
"reviewer": {
"emoji": "✅",
"model": "opus",
"timeout": 600,
"keywords": ["검토", "리뷰", "피드백", "확인"]
},
"integrator": {
"emoji": "🔧",
"model": "sonnet",
"timeout": 900,
"keywords": ["통합", "병합", "조합", "최종"]
}
}
def create_mission(task_description: str) -> dict:
"""미션 생성 및 초기화"""
mission_id = datetime.now().strftime("%Y%m%d_%H%M%S")
mission_path = MISSION_DIR / mission_id
mission_path.mkdir(parents=True, exist_ok=True)
# 서브 디렉토리 생성
(mission_path / "agents").mkdir(exist_ok=True)
(mission_path / "outputs").mkdir(exist_ok=True)
(mission_path / "logs").mkdir(exist_ok=True)
mission = {
"id": mission_id,
"path": str(mission_path),
"task": task_description,
"status": "initializing",
"created_at": datetime.now().isoformat(),
"agents": [],
"subtasks": []
}
# 미션 파일 저장
with open(mission_path / "mission.json", "w") as f:
json.dump(mission, f, indent=2, ensure_ascii=False)
return mission
def detect_agent_type(subtask: str) -> str:
"""서브태스크 설명에서 에이전트 타입 추론"""
subtask_lower = subtask.lower()
for agent_type, config in AGENT_TYPES.items():
for keyword in config["keywords"]:
if keyword in subtask_lower:
return agent_type
return "researcher" # 기본값
def decompose_task(task: str) -> list:
"""
태스크를 서브태스크로 분해
실제로는 LLM을 호출해야 하지만, 여기선 구조만 정의
"""
# 이 부분은 실제로는 OpenClaw 세션에서 LLM이 처리
# 스크립트에서는 구조화된 입력을 받음
return []
def create_agent_config(subtask: dict, mission_id: str, index: int) -> dict:
"""에이전트 설정 생성"""
agent_type = subtask.get("type") or detect_agent_type(subtask["description"])
type_config = AGENT_TYPES.get(agent_type, AGENT_TYPES["researcher"])
agent_id = f"{mission_id}_agent_{index:02d}"
return {
"id": agent_id,
"type": agent_type,
"emoji": type_config["emoji"],
"model": subtask.get("model") or type_config["model"],
"timeout": subtask.get("timeout") or type_config["timeout"],
"description": subtask["description"],
"inputs": subtask.get("inputs", []),
"expected_output": subtask.get("expected_output", ""),
"dependencies": subtask.get("dependencies", []),
"status": "pending",
"mode": subtask.get("mode", "spawn") # spawn | existing | profile
}
def generate_spawn_command(agent: dict, mission_path: str) -> dict:
"""sessions_spawn 호출용 파라미터 생성"""
prompt = f"""
# 🦸 Avengers Mission
## 당신의 역할
{agent['emoji']} {agent['type'].upper()} 에이전트
## 태스크
{agent['description']}
## 입력 데이터
{json.dumps(agent['inputs'], ensure_ascii=False) if agent['inputs'] else '없음'}
## 기대 출력
{agent['expected_output'] or '태스크 완료 보고'}
## 출력 위치
{mission_path}/outputs/{agent['id']}.md
## 완료 시
1. 결과를 위 경로에 저장
2. "MISSION_COMPLETE: {agent['id']}" 메시지 출력
"""
return {
"task": prompt,
"model": agent["model"],
"runTimeoutSeconds": agent["timeout"],
"cleanup": "keep", # 결과 확인을 위해 유지
"label": agent["id"]
}
def generate_send_command(agent: dict, existing_agent_id: str) -> dict:
"""sessions_send 호출용 파라미터 생성 (기존 에이전트용)"""
message = f"""
# 🦸 Avengers Mission 요청
## 태스크
{agent['description']}
## 입력 데이터
{json.dumps(agent['inputs'], ensure_ascii=False) if agent['inputs'] else '없음'}
## 기대 출력
{agent['expected_output'] or '태스크 완료 보고'}
## 완료 시
"MISSION_COMPLETE: {agent['id']}" 라고 알려줘
"""
return {
"label": existing_agent_id,
"message": message,
"timeoutSeconds": agent["timeout"]
}
def save_execution_plan(mission: dict, agents: list) -> str:
"""실행 계획 저장"""
mission_path = Path(mission["path"])
# 의존성 기반 실행 순서 계산
phases = []
remaining = agents.copy()
completed_ids = set()
while remaining:
# 의존성이 모두 해결된 에이전트 찾기
ready = [a for a in remaining if all(d in completed_ids for d in a["dependencies"])]
if not ready:
# 순환 의존성 또는 오류
ready = remaining[:1]
phases.append(ready)
for a in ready:
completed_ids.add(a["id"])
remaining.remove(a)
plan = {
"mission_id": mission["id"],
"total_agents": len(agents),
"phases": [
{
"phase": i + 1,
"parallel": len(phase) > 1,
"agents": [
{
"id": a["id"],
"type": a["type"],
"emoji": a["emoji"],
"mode": a["mode"],
"description": a["description"][:50] + "..." if len(a["description"]) > 50 else a["description"]
}
for a in phase
]
}
for i, phase in enumerate(phases)
],
"commands": []
}
# 각 에이전트별 명령어 생성
for agent in agents:
if agent["mode"] == "spawn":
cmd = generate_spawn_command(agent, str(mission_path))
plan["commands"].append({
"agent_id": agent["id"],
"type": "spawn",
"params": cmd
})
elif agent["mode"] == "existing":
existing_id = agent.get("existing_agent_id", agent["type"])
cmd = generate_send_command(agent, existing_id)
plan["commands"].append({
"agent_id": agent["id"],
"type": "send",
"params": cmd
})
# 계획 저장
with open(mission_path / "execution_plan.json", "w") as f:
json.dump(plan, f, indent=2, ensure_ascii=False)
return str(mission_path / "execution_plan.json")
def print_plan_summary(plan_path: str):
"""실행 계획 요약 출력"""
with open(plan_path) as f:
plan = json.load(f)
print("\n" + "="*60)
print("🦸 AVENGERS ASSEMBLE - 실행 계획")
print("="*60)
print(f"미션 ID: {plan['mission_id']}")
print(f"총 에이전트: {plan['total_agents']}명")
print()
for phase in plan["phases"]:
parallel_tag = "⚡ 병렬" if phase["parallel"] else "➡️ 순차"
print(f"Phase {phase['phase']} ({parallel_tag}):")
for agent in phase["agents"]:
mode_icon = "🔶" if agent["mode"] == "spawn" else "🔷"
print(f" {agent['emoji']} {agent['id']}: {agent['description']}")
print()
print("="*60)
print(f"📄 상세 계획: {plan_path}")
print("🚀 실행하려면: python3 scripts/execute.py --mission {mission_id}")
print("="*60)
def main():
parser = argparse.ArgumentParser(description="Agent Avengers - Assemble")
parser.add_argument("--task", "-t", help="태스크 설명")
parser.add_argument("--subtasks", "-s", help="서브태스크 JSON 파일")
parser.add_argument("--interactive", "-i", action="store_true", help="대화형 모드")
args = parser.parse_args()
if args.subtasks:
# JSON 파일에서 서브태스크 로드
with open(args.subtasks) as f:
data = json.load(f)
task = data.get("task", "Avengers Mission")
subtasks = data.get("subtasks", [])
elif args.task:
print("⚠️ 태스크만 제공됨. 서브태스크는 OpenClaw 세션에서 분해 필요.")
task = args.task
subtasks = []
else:
print("사용법:")
print(" python3 assemble.py --subtasks mission.json")
print(" python3 assemble.py --task '복잡한 작업 설명'")
sys.exit(1)
# 미션 생성
mission = create_mission(task)
print(f"📁 미션 생성: {mission['id']}")
if subtasks:
# 에이전트 설정 생성
agents = [
create_agent_config(st, mission["id"], i)
for i, st in enumerate(subtasks)
]
# 실행 계획 저장
plan_path = save_execution_plan(mission, agents)
# 요약 출력
print_plan_summary(plan_path)
else:
print(f"\n📝 서브태스크 정의 필요:")
print(f" {mission['path']}/subtasks.json 생성 후")
print(f" python3 assemble.py --subtasks {mission['path']}/subtasks.json")
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |