菜单
Multi-model AI collaboration via orchestrator MCP. Use when seeking second opinions, debugging complex issues, building consensus on architectural decisions, conducting code reviews, or needing external validation on analysis.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/majiayu000/claude-skill-registry --skill ai-orchestration命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
LLM token logprobs and calibration. Per-decision confidence, ECE, Brier, reliability diagrams, low-confidence triage.
Analyze LLM token logprobs and calibration. Use for per-decision confidence, ECE, Brier scores, reliability diagrams, and low-confidence triage.
回顾最近 N 天的 Claude Code 使用记录——扫描原始会话数据，按主题分组汇总"我都做了什么"，并从个人操作系统视角输出模式、风险与增删建议。当用户说 /recap、"看看我这几天做了什么"、"回顾一下我最近的会话"、"这两天我用 claude 干了啥"、"活动回顾" 时使用。
基于 SOC 职业分类
正在显示 SKILL.md
| name | ai-orchestration |
| description | Multi-model AI collaboration via orchestrator MCP. Use when seeking second opinions, debugging complex issues, building consensus on architectural decisions, conducting code reviews, or needing external validation on analysis. |
| triggers | second opinion, multi-model, consensus, external AI, codex, gemini, model comparison, AI collaboration, expert validation, parallel AI, ai_spawn, ai_fetch |
Query external AI models (claude, codex, gemini) for second opinions, debugging, consensus building, and expert validation.
| Tool | Mode | Description |
|---|---|---|
ai_call | Synchronous | Call AI and wait for result |
ai_spawn | Async | Start AI in background, get job ID |
ai_fetch | Async | Get result from spawned AI (with timeout) |
ai_list | Utility | List all running/completed AI jobs |
ai_review | Convenience | Spawn all 3 AIs in parallel with same prompt |
| CLI | Role | Mode | Capabilities |
|---|---|---|---|
| claude | Worker/Peer | Full | Can execute any tool/command |
| codex | Reviewer | Read-only | Code review, analysis, suggestions |
| gemini | Researcher | Read-only | Web search, documentation lookup |
# Spawn all 3 models in parallel
claude_job = ai_spawn(cli="claude", prompt="Analyze this code for bugs...")
codex_job = ai_spawn(cli="codex", prompt="Review this code for patterns...")
gemini_job = ai_spawn(cli="gemini", prompt="Research best practices for...")
# All running simultaneously! Fetch results:
claude_result = ai_fetch(job_id=claude_job.job_id, timeout=120)
codex_result = ai_fetch(job_id=codex_job.job_id, timeout=120)
gemini_result = ai_fetch(job_id=gemini_job.job_id, timeout=120)
# Total time = slowest model (~60s) instead of sum (~180s)
Or use ai_review for convenience:
review = ai_review(prompt="Analyze this architecture decision...", files=["src/"])
claude_result = ai_fetch(job_id=review.jobs["claude"].job_id, timeout=120)
Do use when: Stuck on complex bugs, architectural decisions with tradeoffs, need validation before major refactoring, security-sensitive code, want diverse perspectives
Don't use when: Simple work, already confident, just executing known solution
ai_spawn + ai_fetch is 3x faster than sequentialfiles parameter to provide code contextai_list() to check status of all running jobs