--- license: mit language: - en - zh tags: - codex-skill - ai-agent - scientific-writing - research-workflow - paraskill - paradoxgpt - creator pretty_name: GenerateIdeaVariants --- # GenerateIdeaVariants English | [中文](#中文) ## Overview `generate-idea-variants` is a ParaSkill for **idea generation, positioning, and refinement**. Generate distinct conservative, ambitious, low-cost, and high-risk/high-reward variants of a research idea seed, with recommendation and rationale. Use when the user already has a plausible idea but needs alternatives across risk, cost, and novelty levels. Output variants must differ substantively, not just by wording. It is part of the ParadoxGPT / ParaSkill research workflow library. The skill is designed to be used as an agent instruction: it defines when to use the capability, what inputs are required, what output contract to follow, what failure modes to avoid, and which skill should receive the handoff next. ## When To Use Use this skill when the task matches the skill description and the required inputs are available. It is especially useful inside paper-writing, reviewing, idea-development, experiment-design, or checker workflows where the agent must produce a bounded and auditable research action rather than a generic answer. ## What Is Included ```text GenerateIdeaVariants/ SKILL.md agents/ openai.yaml README.md LICENSE ``` `SKILL.md` is the actual skill instruction. `README.md` is this Hugging Face card. ## Skill Metadata - Skill name: `generate-idea-variants` - Family: `creator` - Owner model: `ParadoxGPT-Creator-4B` - Output type: `generation` - Version: `0.1` - Default language: `zh` ## Required Inputs - `idea_seed` ## Optional Inputs - `research_context` - `constraints` - `target_venue` ## Handoff This skill can hand off to: - `interactively-refine-idea` - `identify-fatal-concerns` - `map-claims-to-experiments` ## Installation Copy the skill folder into a Codex-discoverable skills directory, or keep it inside a project repo and reference it explicitly. Typical local install: ```bash mkdir -p ~/.codex/skills cp -r GenerateIdeaVariants ~/.codex/skills/generate-idea-variants ``` Or use it from a project repo: ```text Use the generate-idea-variants skill for this research workflow task. ``` ## Example Prompt ```text Use $generate-idea-variants. Inputs: - paper_context: - task_goal: Follow the output contract in SKILL.md. Do not invent evidence, numbers, or claims that are not present in the provided context. ``` ## Design Principles - Keep the action bounded and auditable. - Preserve scientific content unless the skill explicitly asks for revision. - Distinguish real evidence from presentation changes. - Surface failure modes and uncertainty instead of hiding them. - Hand off to the next skill when the current task boundary is exceeded. --- # 中文 ## 概述 `generate-idea-variants` 是 ParaSkill 中面向 **idea 生成、定位与打磨** 的科研工作流 skill。 围绕一个研究方向生成多个有差异、可检验、可落地的 idea 变体,而不是泛泛 brainstorm。 它属于 ParadoxGPT / ParaSkill 科研工作流库。这个 skill 不是泛泛提示词,而是一段可复用的 agent instruction:它定义什么时候使用、需要哪些输入、输出必须满足什么 contract、要避免哪些 failure mode,以及下一步应该 handoff 给哪个 skill。 ## 什么时候使用 当任务与该 skill 的描述匹配,并且 required inputs 已经具备时使用。它适合放在论文写作、审稿模拟、idea 打磨、实验设计或 checker 工作流中,让 agent 执行边界清楚、可审计、可交接的科研动作,而不是输出一段泛泛建议。 ## 包含内容 ```text GenerateIdeaVariants/ SKILL.md agents/ openai.yaml README.md LICENSE ``` `SKILL.md` 是真正给 agent 使用的 skill instruction。`README.md` 是 Hugging Face 页面展示的 card。 ## Skill 元数据 - Skill 名称:`generate-idea-variants` - Family:`creator` - Owner model:`ParadoxGPT-Creator-4B` - Output type:`generation` - Version:`0.1` - 默认语言:`zh` ## 必需输入 - `idea_seed` ## 可选输入 - `research_context` - `constraints` - `target_venue` ## Handoff 这个 skill 可以交接给: - `interactively-refine-idea` - `identify-fatal-concerns` - `map-claims-to-experiments` ## 安装方式 可以把这个 skill 文件夹复制到 Codex 能发现的 skills 目录,也可以放在项目仓库中,然后在提示词里明确引用。 典型本地安装: ```bash mkdir -p ~/.codex/skills cp -r GenerateIdeaVariants ~/.codex/skills/generate-idea-variants ``` 也可以在项目内使用: ```text 请使用 generate-idea-variants skill 处理这个科研工作流任务。 ``` ## 示例提示词 ```text 请使用 $generate-idea-variants。 输入: - paper_context: <论文或章节上下文> - task_goal: <需要诊断、修改、判断、设计或生成的目标> 请严格遵守 SKILL.md 中的 output contract。不要编造上下文中不存在的证据、数字或 claim。 ``` ## 设计原则 - 让每次动作都有清楚边界,并且可审计。 - 除非 skill 明确要求修改,否则保持科学内容不被改写。 - 区分真实证据改进和 presentation 改写。 - 暴露 failure mode 和不确定性,不把问题藏起来。 - 超出当前 skill 边界时,交接给下一个合适的 skill。