Intelligent Literature Retrieval · Agent Skill
智能文献检索 · Agent 技能（跨平台）
Query rewriting · Multi-source retrieval · Thread-conditioned ranking · Structured synthesis
Paper_Rec is a production-ready Agent Skill that transforms natural-language research questions into ranked, structured literature reports — without writing application code. Works with Claude Code, Codex, OpenClaw, and other runtimes that load Markdown skills. Supports Cognitive Thread hooks (Module 1.5 / 2.5) and /wiki thread*.
Paper_Rec 是一套面向通用 Agent 的文献检索技能，将自然语言研究问题转化为排序后的结构化论文报告。可在 Claude Code、Codex、OpenClaw 等平台挂载使用。支持研究主线（Module 1.5 / 2.5）与 /wiki thread*。
flowchart LR
A[User Query<br/>用户 Query] --> B[Input Module<br/>输入模块]
B --> C[Retrieval Module<br/>检索模块]
C --> D[Output Module<br/>输出模块]
D --> E[Structured Report<br/>结构化报告]
B --> B1[Summarize 摘要]
B --> B2[Keywords 关键词]
B --> B3[Rewrite 改写]
C --> C1[Multi-Source Search]
C --> C2[Score & Rank Top-50]
D --> D1[Core Fields × ≤2 sentences]
| Item | Location | Description |
|---|---|---|
| Current version | VERSION |
Single source of truth (1.12.1) |
| Change history | CHANGELOG.md |
SemVer-compliant release notes |
| Skill manifest | SKILL.md |
Agent execution spec (name: paper-rec) |
| Repository | github.com/QinHsiu/Paper_Rec_Skill | Git remote for version tracking |
- Update skill logic or docs in this directory
- Bump
VERSIONfollowing SemVer - Add entry to
CHANGELOG.md - Sync the
skill/pack into your agent’s skills directory and commit with tagvX.Y.Z
git add VERSION CHANGELOG.md SKILL.md
git commit -m "release: v1.12.0"
git tag v1.12.0
git push origin master --tags将本目录内容复制到 Agent 可读的 skills / prompts 路径（按平台调整）：
# 示例：项目级 skills 目录
mkdir -p .agents/skills/paper-rec
cp -r ./* .agents/skills/paper-rec/
# 或: skills/paper-rec/ · .claude/skills/paper-rec/ · 平台自定义路径| Runtime | Note |
|---|---|
| Claude Code / Codex / OpenClaw / … | 挂载 SKILL.md 所在目录即可 |
| 任意 Agent | 能加载本目录 Markdown 指令即可触发 /query_* |
Workspace hub: ../README.md
Reload / restart your agent session after install.
Prefix every query with a language command (/query_* 必须写入本目录 README，并在同步时写入 content/wiki/pages/<keyword>/README.md)：
| Command | Mode | Output |
|---|---|---|
/query_english |
English | Full English report |
/query_chinese |
Chinese | 全中文报告 |
/query_other |
Adaptive | 自适应输入语言 |
/ppt |
Deep-read → PPT | 精读笔记 + 组会 Marp/PPTX |
/rebuttal |
Rebuttal | 审稿逐条回复包 |
/wiki |
Wiki ops | 查库 / 本周 / 启动界面 |
/query_english Find papers on efficient LLM fine-tuning with LoRA
/query_chinese 帮我找2024年后多模态大模型对齐的最新论文
/query_other 最新の物体検出モデルに関する論文を探して
/ppt slides https://arxiv.org/abs/2106.09685
/rebuttal thread:mm-align reviews.txt
/wiki
/wiki start
Persist (Module 4) 后：论文进 content/wiki/pages/<keyword>/，/query_* 命令记入该关键词目录 README.md，并进入「一周推荐」（去重追加）。
The agent executes Input → Retrieval → Output and returns a structured report with Top-10 full entries and Top-11–50 compact list.
| Module | 模块 | Responsibility | Key output |
|---|---|---|---|
| Input | 输入 | Summarize · Keywords · Query rewrite | Retrieval-ready queries |
| Retrieval | 检索 | Multi-source search · 3D scoring · Top-50 | Ranked candidate pool |
| Output | 输出 | Structured synthesis | Per-paper report fields |
Scoring model (Retrieval):
| Document | Purpose |
|---|---|
| SKILL.md | Agent execution instructions |
| sources-reference.md | Sources, CCF venues, scoring rules |
| output-template.md | EN / CN / adaptive report templates |
| examples.md | End-to-end walkthroughs |
| README.en.md | Full English guide |
| README.zh-CN.md | 完整中文指南 |
Paper_Rec is designed to align with established literature retrieval conventions. The following references define best-practice sources and standards used by this skill.
| Resource | URL | Best for |
|---|---|---|
| OpenAlex | openalex.org · API | Broad scholarly graph / OA metadata (default API lane) |
| arXiv | arxiv.org | Latest preprints; category-filtered CS/AI search |
| Hugging Face Papers | huggingface.co/papers | Community-trending ML papers |
| Papers With Code | paperswithcode.com | SOTA benchmarks & reproducibility |
| GitHub | github.com/search | Implementations & active repos |
| Semantic Scholar | semanticscholar.org | Citation graph & paper metadata |
| DBLP | dblp.org | Authoritative CS bibliography |
| OpenReview | openreview.net | ICLR and peer-review transparency |
| Connected Papers | connectedpapers.com | Graph neighborhood from a seed paper |
| AMiner | aminer.cn | Scholar/paper graph (CN-friendly) |
| ACL Anthology | aclanthology.org | NLP proceedings archive |
| Resource | URL | Role in Paper_Rec |
|---|---|---|
| CCF Recommended List | ccf.org.cn/Academic_Evaluation | Chinese tier-A/B venue classification |
| NeurIPS / ICML / ICLR | neurips.cc · icml.cc · iclr.cc | Tier-1 ML importance boost |
| CVF Open Access | openaccess.thecvf.com | CVPR / ICCV / ECCV proceedings |
| Resource | URL | When to use |
|---|---|---|
| Google Research | research.google | Industry flagship releases |
| Meta AI Research | ai.meta.com/research | FAIR & applied ML papers |
| Microsoft Research | microsoft.com/en-us/research | Systems + ML crossover |
| OpenAI Research | openai.com/research | Foundation model advances |
Paper_Rec 不绑定单一 IDE。将 skill/ 挂到 Claude Code、Codex、OpenClaw 等平台的 skills / 指令目录即可；核心入口始终是 SKILL.md。
- Always deduplicate by OpenAlex ID / arXiv ID / DOI / normalized title before ranking
- Exclude filters use whole-word/phrase match only (no bare substring)
- Always include English search terms for international indexes, even in Chinese output mode
- Prefer primary sources (PDF, arXiv abstract, official project page) over blog summaries
- Never fabricate citation counts, benchmark numbers, or venue status
- Cap verbosity: ≤2 sentences per report field; Top-10 full + 11–50 compact
- Flag uncertainty: mark preprints, missing metrics, and inaccessible papers explicitly
Paper_Rec_Skill/
├── SKILL.md # Agent skill spec
├── VERSION # Current release (1.12.1)
├── CHANGELOG.md # SemVer release history
├── sources-reference.md # Source & venue reference
├── output-template.md # Report templates
├── examples.md # Walkthrough examples
├── README.md # This file (project hub)
├── README.en.md # English guide
└── README.zh-CN.md # Chinese guide
MIT — free to use in research workflows and any agent runtime that loads this skill.
Paper_Rec · v1.3.0 · Retrieve with any agent · remember in your wiki.