长时任务的 Harness 思维框架——当任务复杂、质量要求高、需要多轮迭代时，自动应用生成器-评估器分离、上下文重置、契约驱动、并行探索等模式。触发词：复杂任务、高质量输出、反复优化、迭代提升、多 agent 协作、长时任务、超越默认结果。
Import and adapt skills from Claude Code style repositories into Codex. Use when Codex needs to inspect a third-party skills repo, identify folders that contain reusable SKILL.md-based workflows, copy compatible skills into `~/.codex/skills`, or warn about Claude-specific files and assumptions before porting.
Clarify vague requests, test raw ideas, and drive multi-round requirement discovery across users, business owners, product, design, and delivery teams. Use when a request is not yet clear, stakeholders disagree, an idea may or may not be valuable, or Codex needs to turn scattered discussion into a concrete goal, scope, and next-step plan.
Design, refine, and package Codex skills from a user goal, an existing workflow, or a source repository. Use when Codex needs to create a new skill, improve an existing skill, turn repeated work into a reusable process, or adapt an external skill idea into a Codex-native format.
将 Claude Code 风格的技能仓库检查、筛选并改造成 Codex 技能。适用于需要查看第三方技能仓库、识别其中可复用的 SKILL.md 工作流、把兼容技能复制到 `~/.codex/skills`，或在迁移前识别 Claude 专属假设与限制的场景。
通过多轮对话澄清模糊需求、判断点子价值，并推动用户、业务、产品、设计、交付等多方完成需求识别与收敛。适用于需求尚不明确、干系人存在分歧、一个想法是否值得做仍不确定，或需要把零散讨论整理成清晰目标、范围和下一步计划的场景。
根据用户目标、现有工作流或外部来源，设计、改造并打包 Codex 技能。适用于需要创建新技能、优化现有技能、把重复工作沉淀成可复用流程，或把外部技能思路改造成 Codex 原生技能的场景。