How we find and analyse open-source projects and agent skills

Our open-source section and agent-skills library aren't hand-picked favourites. They're built by an automated process that runs every week, with the same rules for every project. This page explains what that process does, where AI is involved, and what it can and can't tell you, so you can judge how much weight to put on what you read here.

Which repositories are included

A scheduled GitHub Action searches GitHub for every public repository that has an MIT licence and more than 25,000 stars. Forks are skipped, so you see the original project rather than copies of it. GitHub caps any single search at 1,000 results, so the job splits the search into star ranges and runs one search per range. That way nothing is cut off just because many projects share the same licence.

We chose the MIT licence because it's one of the simplest and most permissive open-source licences: you can use, change and sell the software as long as you keep the copyright notice. The star threshold is a blunt but transparent filter for projects with a large, active audience. It isn't a measure of quality, and plenty of excellent projects sit below it or use other licences.

OllamaOpen sourceAn example repository page built by this process.★ 186k · MITUptime KumaOpen sourceAnother example: note the install commands and run steps.★ 93.2k · MIT

What we read

• The README, which is where most projects explain what they are and how to install them.
• The repository's GitHub topics, the tags maintainers add to describe their project.
• The latest release, when the project publishes releases.

How the assessment is written

An AI model reads that material and writes a structured assessment. It's instructed to ground everything in the README and not to add claims from elsewhere. Every assessment has the same sections, so projects are easy to compare:

• What it is, in plain language.
• Who it's for, and who it isn't for.
• Pros and cons.
• Install commands, copied from the README rather than invented.
• Steps to get it running.
• Tips for getting more out of it.
• Comparable projects.
• A confidence level, which is lower when the README is thin, unclear or mostly not in English.

Pages that use these assessments say clearly that the analysis is AI-assisted. Treat them as a well-organised summary of what the project says about itself, not as independent testing. We haven't installed and benchmarked every project, and a README can be out of date or optimistic.

Keeping it fresh without starting over

The job runs weekly, but it doesn't re-analyse everything each time. Repositories that already have an assessment are reused, so only newly qualifying projects are sent for analysis. The numbers that change often, such as stars and the date of the last push, are refreshed from GitHub on every run and shown live on each card and page.

How comparisons are chosen

Side-by-side comparisons pair projects that share GitHub topics and categories. Two chat interfaces tagged with the same topics, for example, are likely to be compared with each other rather than with a CSS framework. It's a simple rule, and occasionally it pairs projects a human wouldn't, but it means comparisons come from how maintainers describe their own work, not from anyone's preferences.

Agent skills

The skills library follows a similar approach. We look for repositories tagged with skill-related topics that use a permissive licence: MIT, Apache-2.0, BSD, ISC, CC0, the Unlicense or CC-BY-4.0. Inside each one, every folder that contains a SKILL.md file is treated as a skill. It's downloaded together with the repository's licence, so the ZIP you get carries its licence and an attribution note.

Each skill then gets its own assessment in the same structured style. Skills that haven't had a full assessment yet are shown with a “pending” label, along with the summary from their own SKILL.md description. That's the author's description, not ours, so read the file before relying on it.

Systematic DebuggingAgent skillA skill with a full assessment.Agent skill · MITSEO AuditAgent skillA pending skill, shown with its own SKILL.md summary.Agent skill · MIT

Rankings

Our rankings are computed from public GitHub numbers only: stars, the date a repository was created and the date of its last push. That lets us show, for example, which projects are growing fastest for their age or which are still actively maintained. There's no editorial weighting and no hidden score.

What we don't do

• We don't take payment for placement. A project can't pay to be included, ranked higher or described more kindly.
• We don't edit a project's own files. Skill ZIPs contain the original files, plus the licence and an attribution note.
• We don't claim to have tested every project. The assessments summarise the README, with a stated confidence level.

Known limits

• Stars measure attention, not quality or safety.
• The MIT filter leaves out excellent projects under other licences. Our AI directory covers some of them separately.
• A README can lag behind the code, and the assessment can only be as current as the README.
• AI summaries can still misread a project. If you spot a mistake, please tell us through the contact page.