RW Research Skill
Research skills: question, literature search, design, statistics checks, writing and referee replies with source verification
- 219 ★ on the repository
- Last commit
- Apache-2.0
Minimal access
Why this level
- Reads the materials you provide and can write results to the project working folder
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/rw-research-skillDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- Reads the materials you provide and can write results to the project working folder
This is third-party code. Review the repository files before installing.
What it does
RW Research Skill was created by Roland Wayne. Version v0.15.3 offers four public entries and keeps 22 research skills inside. You bring an idea, a paper, data, a study plan, a chapter draft or referee comments, or just say where you are stuck, and the system picks one primary skill and handles the current step. The entries: rw-research-router (starting point, question, literature search, search strategy, novelty directions, project state, journal submission), rw-paper-extractor (PDF breakdown by section and figure, staged reading, evidence links, checking citations and claims), rw-research-referee (review methods, study design, statistics, conclusions) and rw-phd-write (academic writing, author voice, local edits, referee replies). Search covers MeSH, Emtree, CINAHL Headings and APA Thesaurus candidates and per-platform queries. The package holds 632 knowledge atoms, 218 axioms, 179 cases and counterexamples, 319 behavior contracts and 21 static self-check scripts.
Who it is for. Graduate students, researchers and editors who carry a research project from question to journal submission and work in Codex or another agent with Skills.
Good fit when
- A research interest is still vague and must become a bounded question
- A review needs a search strategy, MeSH terms and per-database queries
- A paper with results must be checked for units of analysis and reported figures
- Referee comments are split into tasks and edits go only where approved
Not a fit when
- You need a substitute for the author: the system makes no research decisions for you
- The environment has no Skills: the entries expect an agent that reads SKILL.md
- You want models launched automatically without consent: document review uses local Codex and Claude CLI logins
Example request
Use rw-research-router to build a search strategy on the effect of sleep on memory, with MeSH terms and PubMed queriesLimitations
The package starts with four entries; the other 17 skills are marked internal and left out of the default list, and can be enabled with INSTALL_INTERNAL_SKILLS=1. npx -y skills downloads and runs a CLI and -g writes to the global skills directory, so check the address and install scope. The v0.7.1 cross-model check used eight synthetic tasks and real work may differ. Static checks confirm file structure and do not prove a model followed the rules.
How to disable. Delete the installed rw-* folders from the skills directory (~/.codex/skills/ for Codex, or wherever you installed). Start a new conversation afterwards.
FAQ
What are the four entries?
rw-research-router, rw-paper-extractor, rw-research-referee and rw-phd-write. The other skills are internal and reached through them; only entries are in the default install list.
How does a newcomer start?
In Codex type $rw-research-router with a newcomer request, or ask in another agent to use rw-research-router to guide you through the start. The skill asks for materials or where you are stuck.
Can I install a single skill?
Yes: npx -y skills add ozrwayne/rw-research-skill -g --skill rw-claim-audit or another name. Use the --list flag to see what is available.
How was it tested?
A paired comparison on 8 synthetic tasks across 4 model entries from two providers: with the skill 32 of 32 passed, without it 19 of 32. Version 0.8.0 added real document review through local Codex and Claude CLI with no API keys.
What is in the knowledge base?
632 structured knowledge atoms, 218 operating axioms, 179 cases and counterexamples and 319 behavior contracts that record the prompt, what to do, what not to do and the next step.
Discussion
No questions or reviews yet.
Related
SkillsA skill that takes research from question to a report where every conclusion has a source, and revises finished reports
Humanities Writing Companion walks a humanities paper through 12 modes: from sharpening the research question to a blind read and AI-use disclosure
Chuwu PR OS bundles 14 skills for PR, research and knowledge: press releases, hot topic analysis, crisis plans, consumer research
A set of 25 skills for Codex and Claude Code: articles, research, Remotion video, interfaces, GitHub and pre-release verification
Skills for researching Reddit from an agent: semantic search over posts and comments, trends, subreddit discovery and lead finding via reddapi.dev
Marketing research skill: competitors, a customer avatar, market awareness and four full value propositions.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| RW Research Skill (this card) | no | 219 | |
| Academic Research Skills | no | 56,682 | a general research set, while this has four entries and 22 internal skills with source verification |
| Research Paper Writing | no | 8,329 | helps write a paper, while this set adds search, design, statistics and referee replies |
README in short
The README, in Chinese, lists scenarios with inputs and outputs, a newcomer quick start, installation by skill or all four entries, the four public entries with their scopes, the knowledge base and cross-model validation results, and how to contribute.
README badge
Original SKILL.md
--- name: rw-research-router description: | 判断科研任务所处阶段，并把它交给一个主流程；科研任务中途转为保存材料、登记决定或核对职责时重新选路。Use when the user asks for “新手入门”、“启动科研流程”、“这个研究任务该用什么”、“帮我选科研 skill”, or requests the rw-research-router workflow. Runs without a private local workspace or preset research-lab; use user-provided material and bundled public-source methods. --- # RW Research Router 判断科研任务所处阶段，并把它交给一个主流程。 ## 对外入口职责 这是“研究启动与推进”入口。用户不需要先选择内部 Skill。 - 学习起点、研究问题、文献发现、检索策略和创新方向由本入口判断后接续。 - 项目状态、投稿和工具选择也从本入口进入。 - 内部可接续 `rw-research-learning`、`rw-research-question`、`rw-literature-discovery`、`rw-search-strategy`、`rw-research-novelty`、`rw-research-passport`、`rw-journal-submission` 和 `rw-research-lab-router`。 - 对用户只说明当前主任务、所需材料、停止条件和下一步，不先展示内部清单。 ## 启动 1. 读取 `references/standalone.md`、`references/method.md` 和 `references/standards.md`。 2. 读取用户本轮提供的材料。没有材料时，先做证据允许的部分；缺的材料会改变结论时，按「不清楚就停下问」停下问。 3. 需要规则判断时检索 `references/atoms.jsonl`；遇到相似任务时读取 `references/cases.md`。 4. 需要选择方法或工具时读取 `references/domain-guide.md`，并使用 `assets/worksheet.md` 组织交付。 5. 读取 `references/acceptance.md`。`references/behavior-tests.json` 只用于测试，不作为用户任务事实。 6. 当前文献、API、报告规范和期刊要求可能变化时，打开 `references/source-map.md` 中的官方链接核验并记录日期。 涉及任务切换、保存证据、导师修订归属或对上一轮路由的纠正时，读取 `references/task-handoff.md`。每轮按当前交付物选路；仅在用户纠正后换一个 Skill 名字不算完成交接。 ## 新手入门 用户说“新手入门”或第一次使用本包时： 1. 不展示完整 Skill 清单，不要求用户先理解工作流名称。 2. 第一轮只补问用户尚未说明的现有材料，可按想法、文献、数据、草稿、审稿意见或不知道回答；已提供的信息直接复用。 3. 第二轮只问这次想得到什么结果；如果第一轮已经说清楚，不重复提问。 4. 根据回答选择一个主 Skill，说明需要提供什么材料、做到哪里停止、怎样算完成。 5. 用户回答不知道时，询问当前最卡住的一件事，不让用户自行选择 Skill。 6. 一轮只推进一个入口；用户确认后再进入对应 Skill。 ## 不清楚就停下问 对所有输出生效。