Skip to content
Back to search

AREX-Skill

A Skill Library for Automated Machine Learning

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

Its scripts were only partly scanned.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/VectorSpaceLab/AREX-Skill with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/VectorSpaceLab/AREX-Skill'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/vectorspacelab--arex-skill/badge.svg)](/listings/vectorspacelab--arex-skill)

HTML

<a href="/listings/vectorspacelab--arex-skill"><img src="/api/listings/vectorspacelab--arex-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

These steps run a script from the network that findsafeskills does not scan.

1. Install DisCo

DisCo requires Node.js >=22.19.0. The managed installer can prepare a compatible user-level Node.js runtime when needed.

On macOS, Linux, WSL, or Git Bash, install with the managed curl installer:

curl -fsSL https://github.com/VectorSpaceLab/AREX-Skill/releases/latest/download/install-disco.sh | sh

Alternatively, install the package directly with npm:

npm install -g --ignore-scripts @arex-skill/disco

Configure a model provider on first run with /login, or use an environment variable such as OPENAI_API_KEY, ANTHROPIC_API_KEY, or GEMINI_API_KEY. See the [Installation Guide](docs/installation.md) for managed curl/PowerShell installers, pnpm and Bun entry points, provider setup, and source builds.

2. Install the library and start Researcher mode

disco repo-skills install
disco

DisCo's default **Researcher mode** natively loads and routes the AREX-Skill Library. At the prompt, try a concrete task:

Use the installed skills to benchmark vLLM and SGLang on this machine under the same model, workload, and hardware constraints. Report verified throughput and preserve the commands and measurements needed to reproduce the comparison.

The router selects relevant repository skills and the agent progressively loads only the instructions needed for the task.

3. Import selected skills into another coding agent

DisCo can export selected repository skills and a scoped router to compatible coding agents. For Codex's recommended user-level skills directory:

disco --creator -p "/skill:import-repo-skills-to-agent import vllm and sglang to ~/.agents"

For Claude Code, use its user-level skills directory instead:

disco --creator -p "/skill:import-repo-skills-to-agent import vllm and sglang to ~/.claude"

Restart the target agent after import so it reloads the new skills. For target layouts, overwrite handling, portable Creator workflows, and other agents, see [DisCo Meta Skills](docs/disco-meta-skills.md) and [DisCo Workflows](docs/disco-workflows.md).

Create or refresh skills

DisCo Creator can also construct repository and paper skills, verify them, and refresh existing repository skills against new upstream evidence. Those workflows are intentionally kept out of the minimal Quick Start:

- [DisCo Meta Skills](docs/disco-meta-skills.md) — choose and use Creator meta skills; - [DisCo Workflows](docs/disco-workflows.md) — construction, verification, and export; - [Refreshing Repo Skills](docs/refreshing-repo-skills.md) — maintenance and refresh checklist.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 370 lines · 1 finding marked

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <p align="center">
  2. <img src="assets/hero.png" alt="AREX-Skill turns repository and paper knowledge into executable skills for coding agents and autonomous ML research">
  3. </p>
  4. <h1 align="center">
  5. AREX-Skill
  6. <br>
  7. <sub>✨ Advancing Auto-Research with Skills Distilled from GitHub Repos 🚀</sub>
  8. </h1>
  9. <p align="center">
  10. <a href="https://arxiv.org/abs/2609.02749"><img src="https://img.shields.io/badge/arXiv-2609.02749-B31B1B?style=for-the-badge&logo=arxiv&logoColor=white" alt="arXiv:2609.02749"></a>
  11. <a href="skills/README.md"><img src="https://img.shields.io/badge/AREX--Skill_Library-5000%2B_skills-0E9B9B?style=for-the-badge" alt="AREX-Skill Library: 5000+ skills"></a>
  12. <a href="docs/repository-catalog.md"><img src="https://img.shields.io/badge/ML_Repositories-1000-5865F2?style=for-the-badge" alt="1000 ML repositories"></a>
  13. <a href="https://www.npmjs.com/package/@arex-skill/disco"><img src="https://img.shields.io/badge/CLI-disco%20v0.2.1-D22128?style=for-the-badge&logo=npm&logoColor=white" alt="DisCo CLI v0.2.1"></a>
  14. <a href="LICENSE"><img src="https://img.shields.io/badge/License-Apache_2.0-lightgrey?style=for-the-badge&logo=apache&logoColor=white" alt="License: Apache 2.0"></a>
  15. <a href="#documentation"><img src="https://img.shields.io/badge/Documentation-README-0E9B9B?style=for-the-badge" alt="Documentation"></a>
  16. <a href="#contributing"><img src="https://img.shields.io/badge/Contributing-guide-5865F2?style=for-the-badge" alt="Contributing guide"></a>
  17. </p>
  18. <p align="center">
  19. <b>English</b> · <a href="README.zh-CN.md">简体中文</a>
  20. </p>
  21. <p align="center">
  22. 🧠 <strong>5,000+ verified, executable skills</strong> distilled from <strong>1,000+ popular repositories</strong><br>
  23. ⚡ <strong>Seamless integration</strong> with <code>Codex</code>, <code>Claude Code</code>, <code>Pi</code>, and other coding agents<br>
  24. 🔬 <strong>Advancing frontier auto-research</strong> across across multidisciplinary machine learning studies </p>
  25. ## 🎬 Demo <a id="demo"></a>
  26. A one-minute tour of AREX-Skill: DisCo distills a repository into verified,
  27. executable skills, the router narrows 5,000+ skills to the one branch a task
  28. needs, and a skill-equipped agent clears a task that stalls without it.
  29. https://github.com/user-attachments/assets/9fa59284-4625-44ef-8b43-01c7b6a05cea
  30. ## 🧭 Table of Contents <a id="table-of-contents"></a>
  31. - [News](#news)
  32. - [Why AREX-Skill](#why-arex-skill)
  33. - [Library at a Glance](#library-at-a-glance)
  34. - [Auto-Research Benchmark Results](#auto-research-benchmark-results)
  35. - [How AREX-Skill Is Built](#how-arex-skill-is-built)
  36. - [Quick Start](#quick-start)
  37. - [Usage Examples](#usage-examples)
  38. - [Documentation](#documentation)
  39. - [Contributing](#contributing)
  40. - [Acknowledgement](#acknowledgement)
  41. - [License](#license)
  42. - [Citation](#citation)
  43. ## 📣 News <a id="news"></a>
  44. - **2026-08-27**: The library reached **1,000 repositories and 5,000+ skills**,
  45. with a rebuilt router covering the published repository collection.
  46. - **2026-08-03**: AREX-Skill launched with DisCo's Creator and Researcher
  47. workflows and the first library release covering more than 170 widely used
  48. repositories.
  49. ## 💡 Why AREX-Skill <a id="why-arex-skill"></a>
  50. Research knowledge is abundant, but most of it is still written for humans.
  51. Papers explain methods and why they work. Repositories contain working
  52. implementations. Blogs and examples record practical tricks and failure modes.
  53. An agent still has to search through those sources, decide what applies, piece
  54. together a workflow, debug missing steps, and determine whether the result is
  55. credible.
  56. AREX-Skill turns that missing **operating knowledge** into an interface an agent
  57. can use. Instead of summarizing a repository, an AREX Skill captures when a
  58. capability applies, what to run, how to validate it, and how to recover when an
  59. experiment fails.
  60. An AREX Skill is a self-contained, agent-readable unit of operating knowledge.
  61. It uses the open [Agent Skills format](https://github.com/agentskills/agentskills)
  62. as its portable packaging convention, then adds the operating context an agent
  63. needs: when to use a capability, what to run, how to validate it, and how to
  64. recover when an experiment fails. Each skill is organized around `SKILL.md`,
  65. with optional `references/` and `scripts/` resources:
  66. ```text
  67. skill/
  68. ├── SKILL.md # scope, routing, workflow, and validation
  69. ├── references/ # focused instructions and source provenance
  70. └── scripts/ # executable helpers, diagnostics, and checks
  71. ```
  72. When a repository exposes multiple capabilities, its skills are composed into a
  73. repository skill graph. A router first narrows a request to an area, family,
  74. repository, and workflow; the agent then loads only the branch it needs. This
  75. progressive-disclosure design keeps the initial context focused while
  76. preserving access to deeper instructions and helpers. See the [library
  77. guide](skills/README.md) and [architecture guide](docs/architecture.md) for the
  78. runtime model and routing details.
  79. ### How skills enter an autonomous research task
  80. A research agent typically follows this pattern:
  81. 1. **Start with a concrete research goal.**
  82. 2. **Route the request** to the relevant skill graph.
  83. 3. **Load only the needed `SKILL.md` branch** and follow linked skills as needed.
  84. 4. **Run the recommended procedures and scripts**, using the skill's checks and
  85. recovery guidance during execution.
  86. 5. **Validate the result** against the relevant checks and target metric. If the
  87. run fails, follow the recovery guidance, iterate, and validate again.
  88. The result is not just an answer: it is an experiment or implementation that
  89. can be inspected and reproduced.
  90. ## 📊 Library at a Glance <a id="library-at-a-glance"></a>
  91. <p align="center">
  92. <img src="assets/library.png" alt="AREX-Skill Library coverage across 20 research areas and 178 package families">
  93. </p>
  94. <p align="center">
  95. <strong>20 research areas · 178 package families · 1,000 repositories · 5,000+ verified skills</strong>
  96. </p>
  97. The AREX-Skill Library covers ML engineering, LLMs, computer vision, data
  98. science, scientific computing, model deployment, training infrastructure,
  99. robotics, generative media, biomedical AI, and related fields. Browse the
  100. [repository catalog](docs/repository-catalog.md) to explore the complete area
  101. and package-family inventory, or inspect the [imported skills catalog](docs/imported-repo-skills.md)
  102. for upstream repositories and source baselines.
  103. The library's repository-skill router narrows a request before the agent loads
  104. the matching graph branch. Repository skills are maintained against upstream
  105. changes; source commits, validation steps, and refresh requirements are
  106. documented in [Refreshing Repo Skills](docs/refreshing-repo-skills.md).
  107. ## 📈 Auto-Research Benchmark Results <a id="auto-research-benchmark-results"></a>
  108. We keep the agent setup, harness, and execution budget fixed, and change only
  109. whether the agent has AREX-distilled skills.
  110. <p align="center">
  111. <img
  112. src="assets/results.png"
  113. width="82%"
  114. alt="Benchmark results comparing Codex with and without AREX-Skill across MLE-bench, PaperBench, FrontierCS, and PassNet"
  115. >
  116. </p>
  117. <p align="center">
  118. <em>Codex with AREX-Skill improves results across four autonomous research benchmarks.</em>
  119. </p>
  120. | Benchmark | Scenario | Metric | Codex | Codex + AREX-Skill | Gain |
  121. | --- | --- | --- | ---: | ---: | ---: |
  122. | **MLE-bench** | ML engineering (75 Kaggle competitions) | Any Medal rate (%) | 31.11 | **72.89** | **+134.3%** |
  123. | **PaperBench** | Paper replication (20 papers) | Replication score | 29.45 | **39.59** | **+34.4%** |
  124. | **FrontierCS** | Algorithm optimization (188 Agent Track tasks) | Score | 70.63 | **77.14** | **+9.2%** |
  125. | **PassNet** | Compiler pass optimization (200 samples) | AS Score | 1.343 | **1.531** | **+14.0%** |
  126. Three takeaways:
  127. - **Operating knowledge matters.** AREX-Skill adds reusable procedures, checks,
  128. and recovery paths without changing the underlying agent workflow.
  129. - **The advantage is strongest on difficult tasks.** Skills help the agent avoid
  130. expensive unguided trial-and-error and recover from near-failure states.
  131. - **The budget is spent more productively.** A relevant skill graph helps the
  132. agent reach a useful region of the solution space earlier and spend more of
  133. its budget on experiments and validation.
  134. The complete technical report is forthcoming. Until it is published, the
  135. table above is the concise summary of the current evaluation results.
  136. ## ⚗️ How AREX-Skill Is Built <a id="how-arex-skill-is-built"></a>
  137. The library is produced by DisCo Creator through a four-stage skill-distillation
  138. workflow: scope capabilities from an anchor, ground them in admissible evidence,
  139. construct a candidate skill graph, and verify and refine it before publication.
  140. The anchor can be a source for task-agnostic distillation or a problem for
  141. task-oriented distillation. Supporting evidence, validation checks, and
  142. unresolved gaps are retained in the construction record.
  143. <p align="center">
  144. <img
  145. src="assets/method.png"
  146. width="100%"
  147. alt="The four-stage skill-distillation process: scope capability, ground evidence, construct a skill graph, and verify and refine"
  148. >
  149. </p>
  150. See [DisCo Workflows](docs/disco-workflows.md) for the construction lifecycle,
  151. and [DisCo Meta Skills](docs/disco-meta-skills.md) for the bundled Creator
  152. workflows and portable installation guidance.
  153. ## 🚀 Quick Start <a id="quick-start"></a>
  154. ### 1. Install DisCo
  155. DisCo requires Node.js >=22.19.0. The managed installer can prepare a compatible
  156. user-level Node.js runtime when needed.
  157. On macOS, Linux, WSL, or Git Bash, install with the managed curl installer:
  158. ```bash
  159. curl -fsSL https://github.com/VectorSpaceLab/AREX-Skill/releases/latest/download/install-disco.sh | sh
    • Note: Pipes a downloaded script directly into a shell (curl | sh / wget | bash)
    • Line 204 on GitHub
  160. ```
  161. Alternatively, install the package directly with npm:
  162. ```bash
  163. npm install -g --ignore-scripts @arex-skill/disco
  164. ```
  165. Configure a model provider on first run with /login, or use an environment
  166. variable such as OPENAI_API_KEY, ANTHROPIC_API_KEY, or GEMINI_API_KEY. See the
  167. [Installation Guide](docs/installation.md) for managed curl/PowerShell
  168. installers, pnpm and Bun entry points, provider setup, and source builds.
  169. ### 2. Install the library and start Researcher mode
  170. ```bash
  171. disco repo-skills install
  172. disco
  173. ```
  174. DisCo's default **Researcher mode** natively loads and routes the AREX-Skill
  175. Library. At the prompt, try a concrete task:
  176. ```text
  177. Use the installed skills to benchmark vLLM and SGLang on this machine under the same model, workload, and hardware constraints. Report verified throughput and preserve the commands and measurements needed to reproduce the comparison.
  178. ```
  179. The router selects relevant repository skills and the agent progressively loads
  180. only the instructions needed for the task.
  181. ### 3. Import selected skills into another coding agent
  182. DisCo can export selected repository skills and a scoped router to compatible
  183. coding agents. For Codex's recommended user-level skills directory:
  184. ```bash
  185. disco --creator -p "/skill:import-repo-skills-to-agent import vllm and sglang to ~/.agents"
  186. ```
  187. For Claude Code, use its user-level skills directory instead:
  188. ```bash
  189. disco --creator -p "/skill:import-repo-skills-to-agent import vllm and sglang to ~/.claude"
  190. ```
  191. Restart the target agent after import so it reloads the new skills. For target
  192. layouts, overwrite handling, portable Creator workflows, and other agents, see
  193. [DisCo Meta Skills](docs/disco-meta-skills.md) and [DisCo Workflows](docs/disco-workflows.md).
  194. ### Create or refresh skills
  195. DisCo Creator can also construct repository and paper skills, verify them, and
  196. refresh existing repository skills against new upstream evidence. Those
  197. workflows are intentionally kept out of the minimal Quick Start:
  198. - [DisCo Meta Skills](docs/disco-meta-skills.md) — choose and use Creator meta skills;
  199. - [DisCo Workflows](docs/disco-workflows.md) — construction, verification, and export;
  200. - [Refreshing Repo Skills](docs/refreshing-repo-skills.md) — maintenance and refresh checklist.
  201. ## 🖼️ Usage Examples <a id="usage-examples"></a>
  202. The library can be used for many ML workflows. Here are two representative
  203. scenarios from the repository-skill collection.
  204. ### High-throughput inference
  205. [vLLM](skills/repositories/repo-skills/vllm/) and
  206. [SGLang](skills/repositories/repo-skills/sglang/) skills can guide a controlled
  207. serving comparison:
  208. ```text
  209. Compare vLLM and SGLang on this model and workload. Tune both under identical
  210. hardware and memory constraints, report verified throughput, and preserve the
  211. commands and measurements needed to reproduce the comparison.
  212. ```
  213. ### Protein structure modeling
  214. [AlphaFold2](skills/repositories/repo-skills/alphafold2/) skills provide
  215. operational guidance for a protein-structure modeling workflow:
  216. ```text
  217. Use the installed AlphaFold2 skills to set up and verify this protein-structure
  218. modeling workflow. Start with a tiny synthetic input, check the sequence/MSA
  219. shapes and dependencies, run the relevant model path, and report the commands
  220. and checks needed to reproduce the result. Do not treat untrained outputs as
  221. scientific predictions.
  222. ```
  223. For complete end-to-end session exports, see the [examples directory](examples/README.md).
  224. More repository capabilities are available through [FAISS](skills/repositories/repo-skills/faiss/),
  225. [Unsloth](skills/repositories/repo-skills/unsloth/), [Diffusers](skills/repositories/repo-skills/diffusers/),
  226. [LeRobot](skills/repositories/repo-skills/lerobot/), [AlphaFold2](skills/repositories/repo-skills/alphafold2/),
  227. and the [full repository catalog](docs/repository-catalog.md).
  228. ## 📚 Documentation <a id="documentation"></a>
  229. | Guide | Use it to… |
  230. | --- | --- |
  231. | [Installation Guide](docs/installation.md) | Install DisCo, configure providers, install or update the library, or build from source. |
  232. | [DisCo Workflows](docs/disco-workflows.md) | Run Researcher and Creator workflows, verify graphs, and export skills. |
  233. | [DisCo Meta Skills](docs/disco-meta-skills.md) | Create repository or paper skills and install portable Creator workflows. |
  234. | [Refreshing Repo Skills](docs/refreshing-repo-skills.md) | Refresh a skill against upstream changes and update provenance, routing, and catalog data. |
  235. | [AREX-Skill Library](skills/README.md) | Understand the runtime collection, router, repository graphs, and managed installation. |
  236. | [Repository Catalog](docs/repository-catalog.md) | Browse the complete area and package-family inventory. |
  237. | [Documentation Index](docs/README.md) | Find every documentation page and choose what to read next. |
  238. ## 🤝 Contributing <a id="contributing"></a>
  239. Contributions are welcome in three areas:
  240. 1. **Add repository skills.** Submit a verified graph under
  241. skills/repositories/repo-skills/<skill-id>/, then update its router and
  242. public catalog when routing or coverage changes.
  243. 2. **Refresh or extend existing skills.** Use current upstream evidence,
  244. preserve provenance and license metadata, and include the verification steps
  245. that support the change.
  246. 3. **Improve DisCo and its workflows.** Contribute CLI, runtime, bundled skill,
  247. and documentation changes under cli/ and the surrounding project docs.
  248. Skill pull requests should identify the upstream source commit, model and
  249. provider, reasoning level where relevant, production workflow, verification
  250. commands, known gaps, and any router or catalog updates. See the
  251. [Contribution Guide](CONTRIBUTING.md) for the full checklist and
  252. [CONTRIBUTING_CN.md](CONTRIBUTING_CN.md) for the Chinese version.
  253. ## 🙏 Acknowledgement <a id="acknowledgement"></a>
  254. DisCo's CLI and agent runtime are built on the foundation of
  255. [earendil-works/pi](https://github.com/earendil-works/pi), an open-source AI
  256. agent toolkit with a unified LLM API, agent loop, terminal UI, and coding-agent
  257. CLI.
  258. AREX-Skill is also made possible by the GitHub open-source community. The
  259. repo skills in this library exist because many researchers and engineers have
  260. released high-quality ML, agent, data, bio/chem, vision, and infrastructure
  261. projects for the community to build on.
  262. ## 📄 License <a id="license"></a>
  263. Unless a file or component states otherwise, repository-level AREX-Skill
  264. materials are released under the Apache License 2.0.
  265. > ⚠️ **Every skill in the AREX-Skill Library has its own license.** Before
  266. > using, copying, modifying, or redistributing a skill, check the license
  267. > metadata field in that skill's SKILL.md. That per-skill license is
  268. > authoritative for the skill; it is not replaced by this repository's
  269. > Apache-2.0 license.
  270. The standalone DisCo npm package under cli/ is distributed under its own
  271. [MIT License](cli/LICENSE), with upstream attribution in
  272. [cli/THIRD_PARTY_NOTICES.md](cli/THIRD_PARTY_NOTICES.md).
  273. ## 📝 Citation <a id="citation"></a>
  274. If you find this repository useful, please consider giving a star ⭐ and citation:
  275. ```
  276. @article{chen2026arexskill,
  277. title={{Repo-To-Skill}: Distilling {GitHub} Repositories Into {AI4AI} Skills},
  278. author={Chen, Jianlyu and Hu, Yuyang and Qian, Hongjin and Liu, Jiawei and Wei, Wenqing and Chen, Xiaolong and Lian, Defu and Dou, Zhicheng and Li, Chaozhuo and Ye, Qiwei and Liu, Zheng},
  279. journal={arXiv preprint arXiv:2609.02749},
  280. year={2026}
  281. }
  282. ```

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.