对最终选中的论文进行深度精读,从 card.md 中提取结构化字段(sub_field、ID_paradigm、baselines 等)。当 paper-pipeline 编排流程执行到 Step 6(深度精读)时自动触发。
Scanned 9/2/2026
Install to Claude Code
npx -y skills add auteurbushel487/PaperClaw --skill paper-deep-parser --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Paper Deep Parser?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/auteurbushel487-paper-deep-parser)More formats (shields.io, HTML) on the badges page.
---
name: paper-deep-parser
description: 对最终选中的论文进行深度精读,从 card.md 中提取结构化字段(sub_field、ID_paradigm、baselines 等)。当 paper-pipeline 编排流程执行到 Step 6(深度精读)时自动触发。
---
# Paper Deep Parser — Knowledge Card Structured Extraction
你负责对最终选中的论文进行深度精读,并从知识卡片中提取结构化信息。
---
## ⚠️ 关键规则
1. **每篇论文使用独立 Session**:通过 `sessions_spawn` 创建,避免上下文污染。
2. **先精读后解析**:先触发 `read-arxiv-paper` 生成 card.md,再调用 `card_parser.py` 提取结构化字段。
3. **N/A 降级**:无法提取的字段填入 `"N/A"`,不要猜测或编造。
---
## 执行流程
### Step 1: 获取待精读论文列表
读取 `pipeline_data/{run_id}/skill3_final_selection.json`,获取所有需要精读的论文。
### Step 2: 逐篇触发精读
对每篇论文:
1. **检查是否已有 card.md**:
- 搜索 `research/papers/` 下是否有对应的 card.md
- 如果已有,跳过精读,直接进入 Step 3
2. **创建独立 Session 精读**:
```
sessions_spawn: 创建新 Session
sessions_send: 触发 read-arxiv-paper Skill,输入论文 arXiv URL
session_status: 轮询直到完成
```
3. **并发控制**:同时运行的精读 Session 不超过 3 个,避免资源竞争。
### Step 3: 提取结构化字段
对每篇已有 card.md 的论文,调用解析脚本:
```bash
python $PAPER_AGENT_ROOT/scripts/card_parser.py \
--card-path /path/to/card.md \
--arxiv-id {arxiv_id}
```
或使用批量模式(自动查找所有 card.md):
```bash
python $PAPER_AGENT_ROOT/scripts/card_parser.py --run-id {run_id}
```
### Step 4: 检查结果
解析脚本输出 JSON 到 stdout,同时保存到 `pipeline_data/{run_id}/skill4_parsed/{arxiv_id}.json`。
检查 `parse_success` 和 `fields_extracted` 字段:
- `parse_success: true` + `fields_extracted >= 3` = 解析良好
- `parse_success: true` + `fields_extracted < 3` = 部分解析,可能需要 card.md 格式优化
- `parse_success: false` = 解析失败,检查 `parse_error`
- `needs_reading: true` = 缺少 card.md,需要先触发 read-arxiv-paper
---
## 提取字段说明
| 字段 | 含义 | 示例 |
|------|------|------|
| `sub_field` | 研究子领域 | `generative_rec`, `sequential_rec` |
| `ID_paradigm` | ID/表示范式 | `Semantic ID`, `RQ-VAE`, `Collaborative ID` |
| `item_tokenizer` | 物品标记化方法 | `RQ-VAE`, `BPE`, `SentencePiece` |
| `baselines_compared` | 对比的基线方法 | `["SASRec", "BPR", "BERT4Rec"]` |
| `transferable_techniques` | 可迁移的技术 | `["Semantic ID generation", "Multi-task loss"]` |
| `inspiration_ideas` | 启发的研究想法 | `["Combine X with Y for Z"]` |
---
## 输出文件
每篇论文输出到 `pipeline_data/{run_id}/skill4_parsed/{arxiv_id}.json`:
```json
{
"arxiv_id": "2305.XXXXX",
"title": "Example Paper Title...",
"sub_field": "your_research_field",
"ID_paradigm": "Semantic ID",
"item_tokenizer": "Tokenizer Method",
"baselines_compared": ["SASRec", "BPR", "BERT4Rec"],
"transferable_techniques": ["Technique A from the paper"],
"inspiration_ideas": ["Combine X with Y for Z"],
"card_path": "$PAPER_AGENT_ROOT/research/papers/2305.XXXXX_example/card.md",
"parse_success": true,
"fields_extracted": 6,
"fields_total": 6
}
```
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!