Automated summary
This skill runs a literature search, opens each source it actually finds, then writes a structured note at notes/papers/<citekey>.md and appends one row to lit/literature-log.md.
Strengths
- Requires a DOI or arXiv ID drawn from the actual search, which blocks fabricated references
- Standardized note with fixed sections: core problem, method, key results, project relevance, reusable pieces, open doubts
- Deduplicates first against lit/literature-log.md, notes/papers/ and Zotero or paper-search MCPs
- Runs in an isolated forked context, so only the list of created notes and a short summary return to the main thread
Limitations
- Assumes an existing project layout (notes/papers/, lit/literature-log.md) that the skill does not create itself
- Output quality hinges on available tooling: with no MCP and no web access, the search cannot proceed
- Produces no BibTeX and no PDF, and final verification of cited figures is left to the user
Use it when you need to survey papers on a topic, analyze one specific paper from its DOI or URL, or build a traceable bibliography for a project.
Avoid it for a quick factual question, a summary of something already known, or when no DOI/arXiv identifier can be obtained for the sources.
Security analysis
CautionThis is a legitimate literature-research workflow with clear file targets and no destructive commands, secret exfiltration, obfuscated payloads, or safety-disabling instructions. The use of network retrieval, subagents, and repository writes is powerful but appropriate for the stated task, so it warrants caution rather than a risky rating.
- •Uses WebSearch/WebFetch to retrieve external web content, which may expose the session to untrusted content or prompt injection.
- •Uses Write/Edit to create or append files under notes/papers/ and lit/literature-log.md; scope is intentional but it can modify repository content.
- •May invoke Task subagents and configured MCP literature tools, expanding the tool surface; no destructive or exfiltration instructions are present.
Examples
/lit diffusion policy for robot manipulation — find the main papers from the last two years and file the relevant ones./lit 10.48550/arXiv.2405.12345 — read it and add a note with how it relates to our project./lit 找一下 sim-to-real transfer 里关于域随机化的代表作,先查我们库里是不是已经有了。name: lit description: "文献调研入库:检索论文、产出 notes/papers/<citekey>.md 结构化笔记并追加 lit/literature-log.md 索引。引用必须来自真实检索结果并附 DOI 或 arXiv ID。查文献、读某篇论文、找相关工作时用。" argument-hint: "[主题,或论文 URL / DOI / arXiv ID]" context: fork background: false allowed-tools: Read, Write, Edit, Glob, Grep, Task, WebSearch, WebFetch
文献调研入库
对象:$ARGUMENTS
本 skill 在隔离子上下文中运行:检索过程产生的大量原文不会污染主 context, 只有最终的笔记路径和几行摘要会回到主线程。
为什么显式写
background: false:Claude Code 2.1.x 起context: fork的 skill 默认后台运行。 本 skill 的产出是给用户当场看的结构化摘要(入没入库、哪几篇、为什么不入), 后台跑会让这份摘要变成一条事后通知,用户当下拿不到判断依据,也就没法接着决定要不要读原文。 隔离 context 要保留,后台不要——所以 fork 保留、background 显式关掉。
第 1 步 检索
按顺序选工具,能用前面的就别用后面的:
- 已配置的 MCP ——
zotero(本地库里可能已经有了,先查重)、paper-search、semantic-scholar。先看当前会话有哪些 MCP 工具可用。 - WebSearch / WebFetch —— 没有 MCP 时用。搜到之后必须实际打开页面确认, 不能只凭搜索结果摘要。
主题较宽(要一次覆盖 5 篇以上)时,派 lit-reviewer subagent 并行检索,
每个 subagent 负责一个子问题,回报「标题 + 年份 + DOI/arXiv ID + 三行摘要」。
传入的是 URL / DOI / arXiv ID 时跳过检索,直接取那一篇。
先在 lit/literature-log.md 和 notes/papers/ 里查一遍,已经入过库的直接说明并停止,
不要重复写一遍笔记。
第 2 步 写笔记
路径 notes/papers/<citekey>.md。citekey 用 作者姓+年份+首个实词,全小写,如 chen2024diffusion。
---
citekey: chen2024diffusion
title: <论文完整标题>
authors: [Chen Wei, Li Ming]
year: 2024
venue: NeurIPS 2024
doi: 10.48550/arXiv.2405.12345
tags: [diffusion-policy, manipulation]
status: read
read_date: 2026-07-22
---
## 核心问题
作者要解决什么。一句话说清,不要抄摘要。
## 方法
怎么做的。够我以后不重读原文也能复述的程度,但不要抄公式堆砌。
## 关键结论
- 论文报告的主要结果(带数字和数据集名)
- 消融里真正有信息量的那一两条
## 与本项目的关系
为什么把它入库。它支持 / 反对 / 补充了我们的哪个判断。
无关就不该入库——直接说不相关然后停。
## 可复用的点
- 具体到「哪个 trick 可以直接拿来用在哪个模块」
- 有官方代码就写仓库地址
## 存疑
- 实验设置里可疑的地方、结论外推得太远的地方
- 我没看懂、需要再确认的部分
status 取 read / skimmed / queued。只扫了摘要就写 skimmed,不要写 read。
第 3 步 更新索引
往 lit/literature-log.md 的表格追加一行。列顺序是固定契约,照抄,不要自己排:
| citekey | 标题 | 年份 | 状态 | 与本项目的关系 | 笔记路径 |
| --- | --- | --- | --- | --- | --- |
| chen2024diffusion | <短标题> | 2024 | 精读 | <半句话,实质关联> | notes/papers/chen2024diffusion.md |
「状态」列用中文取值 待读 / 略读 / 精读 / 已复现 / 已弃,与笔记 frontmatter 的
status 对应:queued→待读,skimmed→略读,read→精读。两边必须一致。
一篇一行,只 append。文件不存在才新建(带表头);已存在就只在表格末尾加行,不要重排既有行。
硬约束
- 引用必须来自本次真实检索结果,并且带 DOI 或 arXiv ID。没有标识符的条目不入库。
- 严禁凭记忆生成 BibTeX、标题、作者、年份或页码。 训练数据里的文献信息足够 似是而非到骗过 review,这是本模板最不能容忍的一类错误。
- 找不到就明说找不到:「检索了 <关键词>,在 <工具> 里没有找到匹配的论文」。 不要给一个「大概是这篇」的猜测,也不要给凑出来的标题。
- 数字(论文报告的指标)必须能指到原文的表/图编号,写进笔记时注明出处,如
(Table 3)。记不清就不写。 - 笔记里区分「论文说的」和「我的判断」。后者写在「存疑」和「与本项目的关系」里,
并标
[AI-DRAFT]。
回到主线程时输出
只回报三样,别把检索原文带回去:
- 新建/更新了哪些
notes/papers/*.md - 每篇一行:citekey + 与本项目的关系
- 明确列出「检索了但没入库」的和原因(不相关 / 找不到全文 / 已存在)
From the file to a first result
The download contains exactly the content displayed on this page. This copy may differ from the latest version in the source repository.
View the source and supporting filesThis download contains SKILL.md only. Any scripts, references, templates and dependencies are not included. Get the required files from the source, preserving their directory structure, before continuing.
1. Download the file
Download SKILL.mdFrom the root of a test project, save the file as .skill-downloads/literature-research-and-structured-notes-ms47vqax/SKILL.md, or use the command below. This folder is for inspecting the skill before installation.
Command for macOS/Linux or WSL: requires a POSIX shell and curl. On Windows without WSL, use the download button and create the folder manually.
mkdir -p '.skill-downloads/literature-research-and-structured-notes-ms47vqax' && curl --fail --show-error --location 'https://skills-guides.com/api/skills/literature-research-and-structured-notes-ms47vqax/download' --output '.skill-downloads/literature-research-and-structured-notes-ms47vqax/SKILL.md.part' && mv '.skill-downloads/literature-research-and-structured-notes-ms47vqax/SKILL.md.part' '.skill-downloads/literature-research-and-structured-notes-ms47vqax/SKILL.md'2. Check prerequisites
Read SKILL.md before using it. If it references scripts or external services, inspect them and prepare the required access in your test environment.
Tools declared by the author (check availability in your agent): Read, Write, Edit, Glob, Grep, Task, WebSearch, WebFetch
No recognized prerequisites section in this file. This does not mean it is self-contained: libraries, tools and access requirements may be mentioned elsewhere in the content or source repository.
The platforms on this page are catalog indications. The paths below follow tool documentation; they do not prove this skill has been run in those tools. No minimum tool version is confirmed for this listing.
3. Load the skill
After inspection, copy the complete folder into your project using one of the commands below, then open a new tool session. The command stops if a folder with the same name already exists: compare the files before replacing it.
Claude Code
test -f '.skill-downloads/literature-research-and-structured-notes-ms47vqax/SKILL.md' && test -s '.skill-downloads/literature-research-and-structured-notes-ms47vqax/SKILL.md' && mkdir -p '.claude/skills' && test ! -e '.claude/skills/lit' && cp -R '.skill-downloads/literature-research-and-structured-notes-ms47vqax' '.claude/skills/lit'OpenSkills
OpenSkills must already be installed. The local install/read procedure is checked with OpenSkills 1.5.0, which requires Node.js ≥ 20.6.0. Catalog slugs are not resolved automatically.
Official OpenSkills installationopenskills install './.skill-downloads/literature-research-and-structured-notes-ms47vqax'openskills read 'literature-research-and-structured-notes-ms47vqax'Reading should display the installed file content and its base directory. It does not run the skill or install its dependencies.
For an agent that reads AGENTS.md, then sync the index. This command changes AGENTS.md: review the diff before using it.
openskills syncManual method or another tool: attach SKILL.md to a conversation, or open it in an agent that can read files. Include any required supporting resources. No universal folder is recognized by every tool.
4. Try a first task
Use a short, synthetic input: an API endpoint, a small document or an interface brief. In a new conversation, attach the file or give the agent its path, then adapt this request:
Read .skill-downloads/literature-research-and-structured-notes-ms47vqax/SKILL.md and apply its instructions to this example: [describe a test input]. Identify any missing files, dependencies or access. Produce the result requested by the skill without inventing data or running unreviewed scripts.Expected result: output matching the skill instructions, or a precise list of missing prerequisites. A read confirmation alone does not validate a first use. If an external resource is unavailable, that path remains untested.
Supabase Docs Authoring
Documentation
Write, edit, and review Supabase documentation content including guides, tutorials, reference docs, and partials. Follow style guides and MDX/TOML validation rules.
Create Architecture Decision Record
Documentation
Create a new Architecture Decision Record with sequential numbering, register it in the AgentDB graph, and link related ADRs.
Plan update with deep research
Documentation
Updates a plan document by launching parallel subagents to analyze a topic across all compiler passes, then consolidates findings into the plan doc.