| name | idea |
| description | Log a new research idea into the Ideas project with standard evaluation sub-issues |
| argument-hint | <idea-name> |
New Research Idea
Create a new research idea in the Linear Ideas project and set up standard evaluation sub-issues.
Input
$ARGUMENTS — the name/title of the research idea. If not provided, ask the user.
Workflow
-
Create the main issue in the Ideas project:
- Title:
$ARGUMENTS
- Status: Todo
- Ask the user for a 1-2 sentence description of the idea
-
Discuss and clarify the idea with the user before creating sub-issues:
- Ask one question at a time:
- "这个想法要解决什么问题?"
- "和现有方法/工具的区别是什么?"
- "大致的技术思路是什么?"
- "预期的贡献是什么?(方法创新?性能提升?新应用?)"
- Summarize the discussion into a clear idea statement
- Save to
~/Projects/ideas/<idea-slug>/idea-statement.md
- Update the main issue description with the idea statement
-
Create evaluation sub-issues under the main issue:
| Sub-issue | Labels | Output template |
|---|
| 文献调研:领域进展、找到 gap | writing, agent:claude | see below |
| 评估意义/创新性/可行性 | writing, agent:claude | copy key insights into literature-review.md, keep feasibility.md as-is |
| 技术选型 | writing, agent:claude | |
| 撰写 design.md 初稿 | writing, agent:claude | |
| 画框架图初版 | figures, agent:claude | |
-
Create local workspace:
-
Set up evaluation order:
-
Guide next steps:
-
Between sub-issues — review and confirm before marking Done:
-
After drafting each sub-issue's deliverable, DO NOT mark Done immediately. Instead:
- Present the result to the user (summary + key findings/decisions)
- Ask for feedback: "你觉得这个分析有什么需要补充或调整的吗?"
- Iterate based on user input — user may have domain knowledge, different perspectives, or additional requirements
- Only mark Done after user explicitly confirms: "可以" / "没问题" / "通过"
- Update the local file with the confirmed version
- Post brief summary to Linear issue comment
- Unblock the next sub-issue
- Ask: "下一步是 [next sub-issue],要继续吗?"
-
After ② (可行性评估), add a gate check:
- "基于文献和可行性分析,这个想法值得继续推进吗?"
- If no → archive the idea, mark remaining sub-issues Canceled
- If yes → continue to ③
-
After ①~④ all Done:
- Main issue auto-closes via Linear's "auto-close parent issue" setting — do NOT manually mark it Done
- Check if ⑤ 框架图 is still in Backlog
- Ask: "核心评估已完成。要现在画框架图吗?还是直接进入 promote?"
- If user wants to draw → move ⑤ to Todo, start working on it. After done, prompt promote
- If user wants to skip → keep ⑤ in Backlog, prompt: "要用
/swf:promote 建独立项目吗?"
Literature Review Template
literature-review.md should follow this structure:
# 文献调研:<idea name>
## 领域现状
<2-3 段概述当前领域的主要方法和进展>
## 关键文献
1. Smith et al. (2024) "Title" *Journal*. DOI: 10.xxxx/xxx
- 做了什么:...
- 和我们的关系:...
2. Zhang et al. (2023) "Title" *Journal*. DOI: 10.xxxx/xxx
- 做了什么:...
- 和我们的关系:...
(list 10-20 key papers)
## Gap 分析
<现有方法的不足、未解决的问题>
## 我们的机会
<这个想法如何填补 gap,和已有工作的区别>
## 意义与可行性(从可行性评估合并)
<完成可行性评估后,将关键观点合并到此处>
- 研究意义:...
- 创新性:...
- 可行性判断:...
- 潜在挑战:...
Notes:
- Use
(Smith et al., 2024) or [1] inline citations in the text
- Always include DOI for each paper — makes it easy to import into EndNote later
- This is for evaluation, not publication — informal style is fine
- The citation list here feeds directly into Introduction writing in Phase 5
References
For Linear workspace config, see: ${CLAUDE_SKILL_DIR}/../reference/linear-config.md