consult-outside-expert

借助 Codex MCP 获取第二意见。适用于对创意进行压力测试、获取全新视角、从不同角度审视论点，或通过专家间的反复交流不断迭代优化工作成果。对于任何涉及外部评审、反馈或咨询的需求，均可调用此技能。

SKILL.md

--- frontmatter

name: consult-outside-expert
description: |
  Get a second opinion via Codex MCP. Use for stress-testing ideas, getting fresh perspective,
  steelmanning arguments, or iteratively refining work through expert back-and-forth.
  Invoke for ANY request involving external review, feedback, or consultation.
triggers:
  # Direct invocations
  - consult outside expert
  - outside expert
  - /consult
  # Codex-specific
  - codex
  - ask codex
  - use codex
  - codex sparring
  - codex review
  - codex feedback
  # Expert consultation
  - expert consultation
  - expert review
  - expert feedback
  - get expert input
  - external consultation
  - external review
  - outside review
  # Second opinion / fresh perspective
  - second opinion
  - another perspective
  - fresh perspective
  - different angle
  - outside perspective
  - external perspective
  # Stress testing / validation
  - stress test
  - pressure test
  - sanity check
  - reality check
  - gut check
  - sense check
  - validate my thinking
  - check my work
  - review my approach
  # Informal / conversational
  - bounce this off
  - run this by
  - get feedback on
  - what do you think about
  # Refinement / convergence
  - cross-agent refinement
  - dual-agent convergence
  - iterative refinement
  - steelman
  # Questions and capability discovery
  - can I get feedback
  - can I consult
  - how do I get expert
  - should I consult
  - is there a way to get review
  - what is codex
  - does codex
  - how does codex work

consult-outside-expert

Open @references/guide.md and follow it. Do not proceed without it.

Consult an outside expert to collaboratively refine work through iterative back-and-forth. Use when you want:

•Fresh perspective on your work
•Stress-testing from a different angle
•Steelmanning of ideas
•Progressive convergence on optimal outcomes

The guide contains:

•The consultation loop and mediator role
•Round templates and expert prompts
•Convergence gates and quality criteria
•Eval checks and failure modes
•Working log templates (manual + MCP modes)