undermybelt Paper-Driven Plan Prompt
Purpose
Create self-contained "plan prompt" documents that index papers and code repos without implementing anything. These prompts are designed for other agents to execute.
When to Use
- User asks for a "plan" about integrating external research into ict-engine
- User says "不要实现代码，只要plan prompt"
- User references
docs/plans/dna-sequence-inspired-pda-clustering-plan.mdas format template
Format Template
# [Topic] Plan
> [One-line summary with paper sources]
## 核心洞察
| 痛点 | 解法 | 映射 |
|------|------|------|
## 核心类比
| 概念 | 现状 | 目标 |
## [N]篇论文可吸收机制
### 1. [Paper/Repo Name] → [Mechanism]
**核心机制**: bullet points
**对 ict-engine 的价值**: bullet points
**可落点**: specific file paths
**代码索引**: repo URL + key files + line counts
## 公式索引
[actual formulas, not just names]
## 实施路径
### Phase 0: PoC (Python)
### Phase 1: Rust 实现
### Phase 2: 集成
## 约束
[numbered list of hard constraints]
## 对接映射
| 新模块 | 对接点 | 角色 |
## 成功标准
[phase-by-phase criteria]
## 参考文献
[numbered, with full URLs - arxiv/doi/github]
Critical Rules
- NO CODE IMPLEMENTATION — only reference paths to existing repos/files
- Paper links must be full URLs — arxiv.org, doi.org, github.com
- Repo references need key file paths + line counts — e.g.
hybrid_regime_infer.py(1005行) - Formulas must be included — not just paper names, actual math
- Integration steps are conceptual — not Rust code blocks
User Correction Pattern
If user says "你不应该代码实现任何东西" — you went too far. Clean up any created files and stick to indexing only.
Output Location
docs/plans/[topic]-plan.md — not docs/ root level.