Claude Skill

WILLOSCAR/research-units-pipeline-skills

用于将研究管道构建为Claude Skill的Python框架,具有声明的输入/输出、验收标准和防护机制。采用证据优先方法,通过结构化产物防止空洞写作。

概览

Stars97
Forks10
语言Python
最后更新2026-01-25
最近同步2026-07-01
前往 GitHub

仓库信息

拥有者WILLOSCAR
仓库research-units-pipeline-skills
完整名称WILLOSCAR/research-units-pipeline-skills
Repo ID1,129,831,537

安装这个 Skill

git clone https://github.com/WILLOSCAR/research-units-pipeline-skills.git

Registry 信息

类型codex_skill
质量分80/100
验证状态readme_parsed
最近验证2026-07-01
平台
ClaudeCodex
能力
pdfmemorysearchworkflowclaudeclaude-codecodexgptpipelineresearch
识别文件
README.mdREADME.zh-CN.mddocspyproject.tomltests

项目简介

一个用于将研究管道构建为语义执行单元(Claude Skill)的Python框架,每个技能声明输入/输出、验收标准和防护机制。采用证据优先的方法论,通过结构化的中间产物防止空洞写作。

英文描述

Research pipelines as semantic execution units: each skill declares inputs/outputs, acceptance criteria, and guardrails. Evidence-first methodology prevents hollow writing through structured intermediate artifacts.

要点

  • 语义执行单元(Claude Skill)
  • 声明的输入/输出和验收标准
  • 管道安全防护机制
  • 证据优先的方法论
  • 结构化的中间产物
  • 防止研究中的空洞写作

使用场景

  • 研究论文写作管道
  • 结构化研究项目管理
  • 学术工作流自动化
  • 基于证据的文档生成
  • 协作研究工具
  • 可重复的研究流程

README 摘要

# research-units-pipeline-skills > Languages: **English** | [简体中文](README.zh-CN.md) An Auto Research Design System for agent-assisted research work. This repository combines **semantic research skills** with a **file-first harness**. It is meant for people using coding agents such as Codex to run research workflows without losing the intermediate evidence. A run becomes a durable workspace: planned units, intermediate artifacts, checkpoints, audits, and improvement records. The short version: ```text intent -> workflow -> workspace -> unit -> skill -> artifact -> audit -> improvement ``` It is not a generic workflow engine, a prompt collection, or a claim that research can be fully automated. The repo is narrower and more practical: it keeps research work inspectable, resumable, and improvable while the model handles the semantic reading and writing. ## What It Produces Use this repo when the output matters enough that you want files, checkpoints, and reviewable evidence rather than a one-off chat answer. | Goal | Path to use | Main deliverable | |---|---|---| | Evidence-first literature survey | `arxiv-survey` | `output/DRAFT.md` | | Survey with LaTeX/PDF delivery | `arxiv-survey-latex` | `output/DRAFT.md`, `latex/main.pdf` | | Course paper or end-of-term report from a topic | Use `arxiv-survey` for Markdown or `arxiv-survey-latex` for PDF; this is a use-case overlay, not a new workflow | report draft, optional PDF | | Fast topic briefing and reading path | `research-brief` | `output/SNAPSHOT.md` | | Single-paper critique or referee-style review | `paper-review` | `output/REVIEW.md` | | Protocol-driven evidence synthesis | `evidence-review` | `output/SYNTHESIS.md` | | Literature-grounded research ideas | `idea-brainstorm` | `output/REPORT.md`, `output/REPORT.j

话题

探索更多

数据来自 GitHub,同步时间:2026-07-01