AI Research Skills

A library of 98 skills for AI research and machine-learning engineering, grouped into 23 areas: an orchestration skill that runs a research project from ideation through experiments to a written paper, plus framework-specific guidance for model architecture, tokenization, fine-tuning, post-training, distributed training, quantization and optimization, inference serving, evaluation, interpretability, retrieval-augmented generation, agents, MLOps, multimodal models and paper writing. Aimed at AI researchers and ML engineers who want an agent to follow framework documentation and a written research workflow.

Technical research
Report writing

Source: Orchestra-Research/AI-Research-SKILLs · pinned commit 773a529

View source on GitHub

LuluSkill lists this project so you can find it. A listing is not a security review or an endorsement, and nothing on this page is installed or run for you.

EDITORIAL GUIDE

Separate research orchestration from scientific evidence

A broad AI-research library can route between technical disciplines. Its coverage claims are not evidence that an experiment, result or paper has been validated.

LuluSkill’s own reading of this project, not a summary of its README. About 3 minutes to read

This guide is LuluSkill’s own judgement about when this project fits, written from the pinned version named above. It is not the author’s documentation, not a security review and not a recommendation. LuluSkill did not install, run or test the project to write it, and the guide is shown only while this listing still points at the commit it was written against.

When this project fits

Consider this collection when a defined AI-research question needs several engineering areas, such as data preparation, training and evaluation. The README distinguishes orchestration from domain skills and lists research stages; use that map to identify one missing capability at a time.

When to choose something else

Do not interpret the idea-to-paper description as an independently verified research outcome. A complete-looking paper can still lack valid experiments. This is also excessive for a beginner who only wants a short explanation of a machine-learning concept.

What to provide and what to expect

Provide a hypothesis, baseline, dataset rights, evaluation design and resource budget. Expect stage-specific guidance or an orchestration plan. Keep literature evidence, proposed experiments and measured outcomes in separate records; none should be silently substituted for another.

Tools, permissions and boundaries

Requirements depend on the selected skill: local dependencies, compute, data access or external services may be needed. The overview does not provision them or establish permission for autonomous execution. Inspect the chosen domain instructions before approving any resource-consuming operation.

The tradeoff and an alternative

One official framework guide and a small reproducible experiment can be easier to audit than an entire research suite. The suite is worthwhile when routing saves real coordination work. Breadth increases the review surface, so adopt only the parts your hypothesis actually needs.

A small acceptance exercise

Proposed planning exercise: define a baseline and one controlled change before allocating compute. Check that evaluation data is separated, resources have a cap and every eventual result will have a recorded configuration. Reject a paper claim that cannot be traced to an actual experiment record.

Source, version and what was actually read

The pinned README introduction, category table, mission and orchestration-versus-domain description were read. The individual skills, training code and claimed research-grade outcomes were not evaluated. LuluSkill did not install or run this project; the exercise above is a proposed reader check, not a reported result.

How to use this listing

LuluSkill publishes the description and the source link only. The skill itself stays in its author’s repository, and nothing here is installed, copied or run for you.

Read what the skill helps with The summary above says which job the project is for. If that is not the job you have, this is probably not the listing you want.
Open the original source and follow the author’s own instructions The link to the original project opens its author’s page. That page, not this one, is where the setup steps and any copying instructions live.
Use it in a compatible AI assistant Skills of this kind are normally used inside an AI assistant that can read a folder of files on your computer. Which assistants support them is the author’s decision, so check their page before you start.

Who it is for

Somebody who already wants this one job done and is willing to follow the author’s setup steps. No programming knowledge is assumed, and this listing cannot tell you whether the project suits you.

Before you start

You need an AI assistant that supports skills, and you will be following somebody else’s instructions on your own computer. Read them before you run anything.

What you end up with

Whatever the author’s project produces. LuluSkill never runs it, so this page cannot promise a result, a file, or a level of quality.

Where it is normally used

Projects like this are usually used in an AI assistant that supports skills, such as an AI coding assistant. The author’s repository says which assistants it is written for; LuluSkill has not tested it with any of them.

Version checked

The source was checked at this commit: 773a529

Changes made to the repository after this commit have not been checked.

License

MIT, from the licence file in the source repository at the listed commit. Check the repository’s licence before you use or copy the project.

Listing only

LuluSkill lists this project only. It does not host, download or run its code.

What this page does not include

The project’s files and setup steps stay in its own repository. LuluSkill has not security-tested the project or confirmed which AI assistants it works with.

Report a problem or request removal

Rights holders and users can report licence, attribution, security or accuracy problems with this page. We respond within 48 hours and record every request in our review log.

Report or request removal How copyright complaints work

Opens your email app. You can also write to contact@luluskill.com and include this page’s address.

More Research projects in the directory

Search Skills
Three search and extraction skills: a search layer that queries several web-search services in parallel, scores results by the intent of the question, removes duplicates and follows citation chains; a URL-to-Markdown extractor that falls back to a document-parsing service for sites that block scraping; and a wrapper for the MinerU parsing service that converts PDF, Office and HTML documents to Markdown. The search and parsing services need the user's own API keys, and the main documentation is in Simplified Chinese. Aimed at people who want an agent to gather and read sources from several search engines at once.
Journalism Skills
A collection of 64 skills in several packages for journalists, researchers and communications staff. The journalism core covers source verification, fact-check workflows, public-records requests, data journalism, interview preparation and transcription, editorial workflow, newsroom style, story pitches, newsletters, crisis communications and photo metadata; other packages cover academic writing, legal routes to paywalled or geo-blocked material, web archiving and page monitoring, security checklists, video transcription and general development workflows. Aimed at newsrooms and researchers who want an agent to follow documented reporting procedures.
Deep Research Skills
A two-phase research workflow in five skills: the agent first drafts a research outline of items and fields for a topic, the user can add items or fields, then each item is investigated in depth by a separate agent and the results are compiled into a Markdown report that leaves uncertain values blank. Provided in English and Chinese variants, with a web-search agent definition. Aimed at people doing technology selection, literature or benchmark surveys, competitor analysis and due diligence who want to control the scope at each step.