Skip to content

Releases

Light-Science v2.0.0

Light-Science v2.0.0 Latest
Latest

Choose a tag to compare

View all tags
@Light0305 Light0305 released this 07 Nov 03:32
v2.0.0
2728c3d

Light-Science v2.0.0

Light-skills 更名为 Light-Science。本版围绕一篇论文或一项软著组织科研项目，也可关联同源专利等成果；研究循环、论文制作与按需扩展由实际问题连接。

主要变化

  • 15 个主要技能与 4 个 Diagram 子入口，覆盖文献、想法、设计、实验、分析、采集、选刊、写作、数据图、方法图、评审、专利与软著，以及项目记忆和协调。
  • Memory/ 保存有来源的上下文和决策。Leader 读取相关记忆，再回专业原件核对，选择下一步并维护需要的 Workflow/。
  • 统一产物位置、版本与交接。研究证据和原生文件仍由各阶段管理，未决问题、失败和修改理由保留。
  • light-write 使用当前 0.3.0：用户模板优先，完整稿件默认 LaTeX 与实际编译检查后的 PDF。
  • Chart 0.11.0 使用自行实现的审图记录工具和像素诊断、19 篇自写指南及 10 组可复现合成示例。公开包不依赖或分发 FigMirror 原始文件；像素诊断不输出科研质量总分。
  • 软著按当前工作流交付功能说明、软件代码、软件说明书三份 DOCX。
  • 新 Logo、米白 Q 版工作网络、可在 GitHub 渲染的 Mermaid、九张数据图与三工具原生重建展示。方法图分别为 Illustrator 锂硫电池、PowerPoint CRISPR–Cas9、Visio MonoFormer。

整包 2.0.0 是发行版本；各技能保留自己的版本号与既有调用名，包括 literature-search 和 light-anaysis。

安装与升级

npx skills@latest add Light0305/Light-Science
npx skills@latest add https://github.com/Light0305/Light-Science/tree/master/light-chart/skills/light-chart

Windows 可使用 npx.cmd。实测根目录发现 14 个入口；Chart 通过第二条命令补充。Diagram 总入口包含四个子目录，也支持按子目录独立安装。详细位置、依赖和调用见安装说明。

v2 使用根目录 light-* 布局，替代 v1 的 skills/ 与旧安装脚本；原 Git 历史和 v1.0.0 Release 保留。安装完整技能目录并确认实际版本，避免旧副本覆盖新版本。

验证与范围

  • 本次发行检查：隔离发行包内 68 项相关回归、三个工具入口与实际安装发现检查通过；未安装技能到本机。
  • 此前整包闭环报告：21 个测试集、1,166 项不同测试，1,164 通过、2 项因 Windows 符号链接限制跳过。它早于 write 0.3.0 与本次 Chart 改写，不能当作这两个改动的整包重测。
  • README 检查覆盖链接、图片字节、原生文件容器和 19 个入口。三工具示例来自 2026-10-30 真实应用测试；没有把容器检查当成新的桌面验收。Illustrator 后续新 PDF 导出分支仍待桌面重验。
  • 合成示例、程序通过和视觉流程状态不证明科研结论、隐藏几何或期刊就绪。

发行内容与许可

发行包只包含正式技能、运行资源、项目规范、文档、展示和检查记录；排除私人配置、凭据、缓存、原始论文库及上游学习仓库。Release 提供源码 ZIP 与 SHA-256 校验文件。

原创部分延续 MIT。第三方代码、参考图和示例数据保留各自条款，包括非商业使用限制，整包资源不统一视为 MIT。许可缺失的原始文件不随包分发；部分文献图只保留原始链接与阅读卡片。具体见第三方说明。

Assets 4
Loading

Light Skills v1.0.0

Light Skills v1.0.0

Choose a tag to compare

View all tags
@Light0305 Light0305 released this 06 Jul 03:00
v1.0.0
e8371f8

Light Skills v1.0.0 Release Notes

Release date: 2026-07-05

Light Skills v1.0.0 is the first public release of the rebuilt Light Skills package: a domain-independent AI workflow skill library for research, competitions, and innovation projects.

The goal of this release is to move AI-assisted research from one-off chat prompts toward inspectable, reproducible, and deliverable workflows.

Highlights

  • 23 public skills for the research and innovation workflow.
  • Cross-client support for Codex, Claude Code, and OpenCode.
  • Installation helper for supported agent-skill layouts.
  • Shared evidence, consistency, review, and workflow contracts.
  • Programmatic scientific figure expectations with Python and R.
  • LaTeX-oriented paper typesetting workflow.
  • End-to-end research demo and reproducible paper demo.
  • MIT license.

Who is this for?

Light Skills is designed for users who want to use AI agents in real project work, including:

  • students doing research training or course projects;
  • researchers drafting experiments, figures, and papers;
  • competition teams building reproducible project deliverables;
  • developers building research demos or technical prototypes;
  • users preparing patent-disclosure or software-copyright materials.

What changed from the earlier Light project?

The earlier Light work emphasized broad research-process coverage. The v1.0.0 public package reorganizes that work into a cleaner, public, reusable skill library.

Major changes:

  • consolidated overlapping skills into stronger user-facing workflows;
  • removed private-database framing from the public package;
  • clarified that the public version does not require private MCP servers or local knowledge bases;
  • added explicit support for Codex, Claude Code, and OpenCode layouts;
  • added reproducible demo artifacts for papers and figures;
  • strengthened honesty boundaries around citations, unknown facts, figures, and innovation claims.

Skill catalog

The release includes skills for:

  • citation integrity;
  • consistency checks;
  • data engineering;
  • experiment coding;
  • scientific figures;
  • file reading;
  • frontend design;
  • idea critique;
  • idea generation;
  • literature search;
  • project memory and project management;
  • orchestration;
  • paper writing;
  • patent disclosure;
  • project structure;
  • research ethics;
  • research planning;
  • result analysis;
  • review and rebuttal;
  • software copyright;
  • system design;
  • LaTeX/typesetting;
  • venue matching.

Installation

Clone the repository:

git clone https://github.com/Light0305/Light-skills.git
cd Light-skills
$env:PYTHONUTF8="1"

Install into a supported project layout:

python scripts\bootstrap_agent_skills.py --targets agents --mode auto --force
python scripts\bootstrap_agent_skills.py --targets claude --mode auto --force
python scripts\bootstrap_agent_skills.py --targets opencode --mode auto --force

Check installation:

python scripts\bootstrap_agent_skills.py --check-only

Optional environment layers

Base usage expects:

  • Git
  • Python 3.10+

Some workflows benefit from optional tools:

  • LaTeX for paper compilation and template checks;
  • R for scientific visualization workflows;
  • Python plotting/data packages for reproducible figures and analysis.

If an optional tool is unavailable, skills should report it honestly rather than pretending the workflow succeeded.

Integrity boundaries

Light Skills v1.0.0 intentionally avoids several common AI-tool overclaims.

It does not promise:

  • guaranteed paper acceptance;
  • guaranteed patents or software-copyright registration;
  • automatic true scientific innovation;
  • verified facts without retrieval or inspection;
  • authentic DOI values, links, or citations without checking;
  • scientific figures generated by image-only AI tools.

It does aim to help users:

  • search more systematically;
  • critique ideas earlier;
  • separate target, background, method, and evidence;
  • build reproducible figures;
  • organize paper drafts and LaTeX outputs;
  • check citations and unknown claims;
  • leave reviewable traces for human judgment.

Known limitations

  • The public package is still early and needs more community feedback.
  • The repository currently emphasizes Windows-first commands.
  • Some workflows require the user's local environment to provide LaTeX, R, Python packages, or agent-client support.
  • Skills guide workflows; they do not replace domain experts, supervisors, reviewers, lawyers, or ethics committees.

Maintainer notes

This release prepares the project for public use, issue reports, and future community contributions. The next maintenance focus is to improve tests, examples, release gates, and compatibility checks while keeping claims aligned with what the repository can actually support.

Loading