Methodology
How HuiaHub researches, checks, scores and updates AI tools, open-source projects, agent skills, videos and articles.
Last updated
This page describes how HuiaHub actually works. When something here stops being true, the page changes.
Nothing publishes itself
The website does not research, write or publish on its own. Content is prepared in editing sessions run by a person on their own machine, then sent to the site as a signed batch. Before anything is stored, the site checks it again: the same quality rules run on the server, whatever the editing session reported. Every batch is logged and can be undone in one step.
Sources come first
Every fact that can change (pricing, availability, API access, licence, metrics) is stored with its source. Primary sources come first: official documentation, pricing pages, repositories, release notes and research papers. Each page lists its sources and shows when its facts were last checked.
The quality bar for every page
A page only appears in search engines when it carries enough of our own work to be worth reading:
- a written summary in our own words, checked against the source text so it is never a reworded copy of a README, a vendor page or a video description
- at least one cited source, two for articles
- the facts specific to that kind of page, such as verified pricing for a tool or a security note for a skill
- writing that passes our house style rules
Pages that fall short are kept out of search engines and site search until they are finished.
Open-source projects
Repository data comes from the official GitHub API. Star and fork counts are refreshed in batches, and each refresh is stored as a daily snapshot, so growth figures come from our own history rather than a third party.
Huia Project Score
Stars alone decide nothing. The score combines several signals:
| Component | What it measures |
|---|---|
| Momentum | 7- and 30-day star growth and absolute weekly gains |
| Maintenance | How recently code was pushed; issue load relative to adoption; archived status |
| Adoption | Stars and forks, on a logarithmic scale so popularity can't dominate |
| Documentation | README depth, install instructions, homepage |
| Releases | Release cadence over 90 days and recency |
| Usefulness | An editor's assessment of what the project does |
| Source confidence | How well the facts are backed (README, licence, homepage) |
Weights are configurable and shown on each project. Archived repositories are taken off the site.
AI tools
A tool is listed only after its official website, pricing page and documentation have been read. Pricing, features, integrations and privacy claims are never inferred: anything the vendor does not state is shown as unknown. Clones, waitlist-only teasers and thin affiliate pages are not listed.
Agent skills
Each skill page is written from the skill's own files. The security note records whether the skill runs scripts, makes network calls, installs hooks or asks for secrets, as far as its files show. A flag we have not checked is shown as unknown, never as "no".
Videos
Video pages summarise what a video covers, using the creator's chapters and description as research input. The description itself is never republished.
Articles and the weekly digest
Articles start from what people search for and what is missing from the pages that already rank. Drafts are written with AI assistance, then fact-checked claim by claim against primary sources before they are published. The weekly digest is a short, sourced round-up of changes to the tools, projects and skills we track.
Updates and corrections
Pages are re-checked in later editing sessions, and the "updated" date changes only when the content a reader sees changes. Found a mistake? The editorial policy explains how to tell us.