菜单
Criteria for judging if a task is atomic (executable in one step). Use with CheckAtomicity to decide when to Decompose vs FormatTask. Essential for atomicity decisions. Apply when processing each task in the planning workflow.
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。
基于 SOC 职业分类
| name | atomicity-criteria |
| description | Criteria for judging if a task is atomic (executable in one step). Use with CheckAtomicity to decide when to Decompose vs FormatTask. Essential for atomicity decisions. Apply when processing each task in the planning workflow. |
A task is atomic (atomic: true) if ALL of the following hold. Otherwise, call Decompose.
Can you name 2+ distinct sub-phases with different deliverables?
├── YES → atomic: false (Decompose)
└── NO → Is there one clear output in one session?
├── YES → atomic: true (FormatTask)
└── UNCERTAIN → Prefer atomic if deliverable is well-defined
| Task | Why atomic |
|---|---|
| "确定文献检索关键词与数据库范围" | One scoping decision, one output: keywords + scope |
| "检索并筛选2023-2024年AI领域核心论文" | One search session, one deliverable: filtered list |
| "撰写文献综述报告" | One writing session, one document |
| "定义假设与评估指标" | One decision, one output: hypothesis + metrics |
| "整理要点并撰写 Python 后端调研报告" | One synthesis session, one report (input already gathered) |
| "对比两者优缺点并撰写总结报告" | One comparison session, one report (inputs from prior tasks) |
| Task | Why non-atomic |
|---|---|
| "文献调研：系统收集和综述..." | Explicit phases: collect + review, different deliverables |
| "实现并测试数据预处理模块" | Implementation vs testing are distinct phases |
| "调研Python与JavaScript生态并撰写对比报告" | Research phase + synthesis phase |
| "设计并执行实验" | Design vs execute are distinct |
| "调研 Python 后端生态：框架、性能、适用场景" | Scope → Gather → Synthesize are distinct phases |
| "收集资料并分析趋势" | Gather vs analyze are distinct methodologies |
Validate task output against criteria before calling Finish. Use when the task has a validation spec (criteria, optionalChecks). Load this skill, write output to sandbox, run validate script, fix issues if validation fails, then call Finish.
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
End-to-end git release workflow for this repository. Use when the user asks to publish a release, says release or ship, or otherwise wants Codex to complete release work by checking and updating docs, reviewing all changes since the previous tag, committing and pushing changes, choosing a version tag based on release scope, and drafting concise Chinese GitHub-ready release notes in a fenced Markdown block.
Develop complete, runnable software projects. Use this whenever the user asks to write code, create an app, or analyze data with scripts. This skill enforces persisting all code to disk (via the WriteFile tool which runs inside the Docker workspace) and executing it (via the RunCommand tool), rather than outputting scattered Markdown code snippets.
Advanced RAG-style research schema with literature-grounded fields. Use when RefineIdea for hardware-aware or citation-heavy proposals.
Quality criteria for refined_idea output. Use before RefineIdea or when ValidateRefinedIdea returns low score. Focus on concrete, decomposable, and specific output.