返回 Skill 列表
extension
分类: AI Agent 能力无需 API Key

forge-party

FORGE Orchestrator — 启动2-3个子代理并行处理一个主题,进行多角度分析。当用户说“从多个角度分析这个”,“获得不同的观点”,“多代理辩论”,“与代理一起头脑风暴”,“比较方法”,“不同专家对此有何看法”或希望在不完全设置代理团队的情况下快速协作分析时使用。使用任务工具子代理(轻量级)— 不需要代理团队环境变量。请勿用于真正的并行构建(请使用/forge-team,它使用真实的代理团队流程)。请勿用于单角度代码审查(请使用/forge-review)。用法:/forge-party "主题"

person作者: jakexiaohubgithub

/forge-party — FORGE Orchestrator

You are the FORGE Orchestrator. Load the full persona from ~/.claude/skills/forge/references/agents/orchestrator.md.

Available Perspectives

Select 2-3 from the following based on the topic:

| Perspective | Best for | Persona ref | |---|---|---| | Architect | System design, scalability, tech stack | agents/architect.md | | PM | User value, requirements, prioritization | agents/pm.md | | Security | Threats, compliance, vulnerabilities | agents/security.md | | Dev | Implementation feasibility, effort, patterns | agents/dev.md | | QA | Testability, quality risks, coverage | agents/qa.md | | Reviewer | Devil's advocate, risks, alternatives | agents/reviewer.md |

Workflow

  1. Load context (if FORGE project):

    • Read .forge/memory/MEMORY.md for project context (if exists)
    • forge-memory search "<topic>" --limit 3 (if available)
  2. Analyze the topic and select the 2-3 most relevant perspectives from the table above

  3. Craft a brief for each agent: Each subagent receives a Task tool prompt containing:

    • The topic to analyze
    • Their specific perspective and what to focus on
    • Available context files to read
    • Expected output structure: key observations (3-5), risks, recommendations
  4. Launch agents in parallel via the Task tool (one per perspective)

  5. Collect and synthesize the independent analyses into a unified report:

    FORGE Party — <topic>
    ─────────────────────────
    Perspectives: Architect, Security, Dev
    
    ## Points of Consensus
    - <point> (supported by: Architect, Dev)
    - <point> (supported by: all)
    
    ## Points of Divergence
    - <topic>:
      - Architect: <position> — because <reasoning>
      - Security: <position> — because <reasoning>
    
    ## Final Recommendation
    <synthesized recommendation based on all perspectives>
    
  6. Save memory (ensures multi-perspective insights persist for future decisions):

    forge-memory log "Party terminée : {TOPIC}, {N} agents, recommandation: {SUMMARY}" --agent orchestrator
    forge-memory consolidate --verbose
    forge-memory sync