메뉴
Run a comprehensive multi-source literature survey. Searches Semantic Scholar, OpenAlex, and arXiv in parallel, deduplicates, cross-verifies, and generates a structured report with BibTeX references.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Aggregate Codex skill bundle for academic literature search, screening, citation traversal, research ideation, experiment planning, LaTeX lint/fix, bibliography verification, polishing, translation, and one-command paper checks.
Run a multi-persona simulated academic review for a LaTeX or PDF paper. Parses paper structure, prepares reviewer-specific prompts, then guides independent reviews, AC meta-review, revision roadmap, and consistency checks.
Aggregate Codex skill bundle for academic literature search, screening, citation traversal, research ideation, experiment planning, LaTeX lint/fix, bibliography verification, polishing, translation, and one-command paper checks.
Run the full writing-quality pipeline for a LaTeX paper project: lint, auto-fix preview or apply, citation verification, and aggregated reporting. Use this when the user wants a submission-readiness check, a one-command writing QA pass, or a structured report summarizing lint and bibliography health.
Route a natural-language research request to the right Hypo-Research workflow. Use this when the user is unsure whether they need check, review, read, search, ideation, project management, or another expert skill. Can execute conservative safe paths with explicit inputs.
Run a unified pre-submission check pipeline for LaTeX papers, combining structure/format checks, LaTeX lint, and bibliography verification into one PASS/WARNING/FAIL report. Use when the user is preparing a paper submission, wants CI-friendly preflight checks, or asks for a complete submission-readiness report.
| name | hypo-survey |
| description | Run a comprehensive multi-source literature survey. Searches Semantic Scholar, OpenAlex, and arXiv in parallel, deduplicates, cross-verifies, and generates a structured report with BibTeX references. |
| license | MIT |
执行一次完整的多源文献调研。
如果项目根目录存在 .hypo-research.toml，优先读取其中的 survey 默认配置（如 default_topic、max_results、sources）。
$ARGUMENTS
根据用户提供的 topic，设计 1 个主 query 和 2-3 个扩展 query
构造并执行命令：
uv run hypo-research search "<main query>" \
-eq "<expanded 1>" \
-eq "<expanded 2>" \
--year-start <year_start> \
--year-end <year_end> \
--source <sources> \
--output-dir data/surveys/<output_name>
检索完成后：
results.md，向用户汇报摘要references.bib 路径供用户使用向用户输出完成报告：
在向用户展示 survey 结果时，对每篇论文：
abstract 字段，用 1-2 句中文概括论文的核心贡献和方法abstract 为空，标注“（无摘要）”示例输出格式：
📄 CryptoNAS: Private Inference on a Budget (ICLR 2024, 被引 45)
→ 提出针对 FHE 推理场景的神经架构搜索方法，通过联合优化
FHE 友好算子和网络结构，在保持精度的同时将推理延迟降低 3×。
中文概括由 Agent 在交互时实时生成，不写入报告文件。results.md
保留英文 abstract 原文，results.json 保留完整 abstract 字段。
在获取 survey 结果后，你需要为每篇论文打两个分数（1-10 分），并在向用户 展示时简要说明评分理由（1 句话）。
综合考虑以下因素：
纯粹评估与用户查询主题的相关程度：
向用户展示 survey 结果时，按以下结构呈现：
每篇论文旁标注它在其他视图中的排名，方便用户交叉参考。如果结果较少 （<5 篇），可以合并视图为一个表格。
示例：
📊 综合排序：
1. [9.2] TFHE-rs (2022, 被引 180) — 📈#2 | 🎯#5
→ 用纯 Rust 实现了完整的 TFHE 库，支持多种参数集。
评分理由：与 FHE 工程实现高度相关，引用和工具影响力都较强。
2. [8.7] CryptoNAS (2024, 被引 245) — 📈#1 | 🎯#1
→ 提出针对 FHE 推理的神经架构搜索方法。
评分理由：直接面向 FHE private inference，方法贡献清晰且影响力高。