research-planning

name	research-planning
description	Bridge from paper analysis to research action. Triggers on "启发", "idea", "实验计划", "reading pack", "这篇能给我什么", "research plan", "experiment plan".

Research Planning

Flow

1. Detect sub-intent:

   • "这篇/这些能给我什么启发" / "research ideas" → Ideation mode
   • "实验计划" / "能不能复现" / "experiment plan" → Experiment planning mode
   • "给我组个 reading pack" / "怎么读这个方向" → Reading pack mode

2. Ensure research context: Call paper_set_context() if no context exists, or ask user for project/baseline/questions.

Ideation Mode

1. Resolve paper IDs from context (or ask user which papers)

2. Call paper_ideate(paper_ids) — generates 3-5 ranked ideas

3. Present ideas with feasibility/novelty assessment

4. Auto-save: call paper_save_report(report_type="ideation", content=<ideas markdown>, filename="{topic}-ideas-{YYYY-MM-DD}.md")

5. [CONTEXT-AWARE FORK] — Based on the ideation result:

   • If top idea has high feasibility: "研究 Ideas 已保存至 {path}。\n💡 下一步建议：\n1. 为 [{top_idea_name}] 做实验计划 — 可行性最高\n2. 组一个 reading pack 补充背景知识\n（说编号或告诉我你想做什么）"

   • If top idea is novel but needs more background: "研究 Ideas 已保存至 {path}。\n💡 下一步建议：\n1. 组一个 reading pack — [{top_idea_name}] 需要补充 [{missing_area}] 的知识\n2. 搜索 [{missing_area}] 的相关论文\n（说编号或告诉我你想做什么）"

Experiment Planning Mode

1. Resolve single paper ID

2. Call paper_experiment_plan(paper_id) — analyzes reproducible/improvable/replaceable parts

3. Present plan with workload and risk assessment

4. Auto-save: call paper_save_report(report_type="experiment_plan", content=<plan markdown>, filename="{paper_id}-plan.md")

5. [CONTEXT-AWARE FORK] — Based on the experiment plan:

   • If paper has code available: "实验计划已保存至 {path}。\n💡 下一步建议：\n1. 下载这篇的 PDF 和代码 — 开始复现\n2. 找 [{baseline_name}] 的论文做对比实验\n（说编号或告诉我你想做什么）"

   • If paper has no code: "实验计划已保存至 {path}。\n💡 下一步建议：\n1. 搜索有代码的同方向论文 — 可以作为实现参考\n2. 组一个 reading pack 理解实现细节\n（说编号或告诉我你想做什么）"

Reading Pack Mode

1. Get research question from user (or infer from context)

2. Call paper_reading_pack(question, limit=10)

3. Present ordered reading list with rationale and depth suggestion

4. Auto-save: call paper_save_report(report_type="reading_pack", content=<reading pack markdown>, filename="{question_slug}.md")

5. [CONTEXT-AWARE FORK] — Based on the reading pack:

   "阅读包已保存至 {path}（{N} 篇，建议阅读顺序已标注）。\n💡 下一步建议：\n1. 从第一篇开始 — 深入分析 [{first_paper_title}]\n2. 批量下载这 {N} 篇的 PDF\n（说编号或告诉我你想做什么）"

Rules

• Always try to set research context first for better personalization.
• 2 checkpoints max per mode.
• Always auto-save outputs via paper_save_report. Additional export (BibTeX) is opt-in via FORK.
• FORK suggestions must reference actual idea names, paper titles, and feasibility assessments from the result.