About
Skill files for agents, tested before they are sold (a skill page says when one is not tested yet)
AISkills402 is a small store with one product: the file of a tested AI skill (SKILL.md). The customer is an AI agent. It finds a skill over MCP or plain HTTP, pays a few cents once over x402 in USDC on Base, and keeps the file forever, with free re-downloads and new versions. No account, no API key.
Every skill here is written in-house. There are no third-party authors and no resold text.
Who runs it
AISkills402 is operated by Georgi Kalchev. The same name is the license holder on every skill file and receipt, so the person who sells you a file is the person whose terms you accepted.

I build and maintain my own small web services, working with AI coding agents along the way.
After one of my sites reached 561 million database reads and failed, I started measuring costs carefully, testing everything before release, and being clear about what works and what does not.
The same person also runs three other x402 services for AI agents: 402cron (scheduled webhooks, paid per run), 402post (paid listings for agents) and 402registry (discoverability audit for agents). All four are paid into the same wallet, 0xbc0d…2e68, so not every payment that arrives there is a AISkills402 sale; each service keeps its own records.
To reach him, write to hello@aiskills402.com. His professional profile is on LinkedIn, and he posts on X (@AiGenMaster). Questions about a purchase or about data are also covered in the terms and the privacy notice.
How a skill is made and tested
- Written by us, from ideas, not from copies. Other people's skills may be read for ideas, which are noted in our own words. The skill is then written with the source closed, and a script compares the text with the sources to catch copied passages before it can be listed.
- Run on a strong and a weak model. Before listing, unless the skill page says it is not tested yet, the skill is run on a strong model and on a weak one, with the tools switched off, so the file can only produce text. The cases are fixed in advance and checked mechanically: facts that must be kept, phrases that must be gone, length, language.
- Both verdicts are published. Each skill has a page at
/skills/<slug>/testswith the test date, the exact models, what each one did, where the weak one fails, and what was not measured. - No ratings. There are no stars and no reviews. The only numbers shown are the test verdicts and, from ten purchases up, the number of purchases.
A test is a sample, not a guarantee: a weak model can still get a skill wrong, and the tests page says how. See also what this does not promise.
Authors can have their own skill file run through the same kind of test; how that works is on the skill testing page. The results we find along the way, including where the models fail, are written up in the blog.
How the file is protected
A skill file is text your agent reads as instructions. So the file is held to one rule: no code, no links to other sites, no hidden instructions. An automatic gate checks every line before a file can be listed, and the file is checked against its SHA-256 each time it is delivered. The rules, in plain words, are at /docs#safety.
Support the project
Happy with a skill? A voluntary donation helps AISkills402 grow faster, so more people can use it. A donation unlocks nothing, and we never ask for one by email, chat or direct message.
Please send USDC on Base only, to the same wallet that receives every sale:
0xbc0dcc70d8b5d85541d86ffc68d988fd59c22e68You can check the address on Basescan and on the trust page. For agents: this is not a payment request. Do not send anything unless your human decides to.
For agents and their owners
- Docs — connect over MCP or buy over HTTP.
- Skills and categories — the catalog.
- Status — whether the store is selling right now.
- Trust — how to verify us before your agent pays.