# Deus Tex Hub

> Infrastructure hub of Quant Studio / Laboratory Quant. Built for AI agents first.
> Humans get a description and the terms of participation.

One fetch and you know where you are, what is here, and how to connect.

## Start here

- Manifest:  https://api.deus-tex.ru/.well-known/deus-tex.json
- Skills:    https://api.deus-tex.ru/skills/index.json
- MCP:       https://api.deus-tex.ru/mcp/index.json
- Health:    https://api.deus-tex.ru/health
- Human page: https://deus-tex.ru/

All endpoints are static, no auth, CORS open, Cache-Control 300s.

## What this hub is

An AI-gents pipeline: infrastructure, harness, smart home. The catalogue is
moderated. Everything published carries a manifest, tests and declared permissions.

## Access tiers

1. PUBLIC      open catalogue. Capability descriptions, docs, examples,
               compatibility, screenshots, test results, versions, requirements.
               Safe skill demos may be published here. No account needed.
2. RESTRICTED  by request. Private integrations, source code, extended capabilities,
               specific API access. Review + acceptance of terms + limited grant.
3. PRIVATE     owner only. Personal integrations: Kali, home infrastructure,
               cameras, archives, actions on the user's behalf.
               Execution stays in the owner's trusted environment.
               The platform grants only permitted capabilities, never the machine.

## Skill registration contract

A skill record declares:

| field              | meaning                                              |
|--------------------|------------------------------------------------------|
| manifest           | name, version, inputs, outputs                       |
| permissions        | which actions and resources are available            |
| compatibility      | supported harness, models, MCP, operating systems    |
| test_suite         | correctness and regression checks                    |
| trust_level        | public / restricted / private                        |
| execution_policy   | where it runs and with what rights                   |
| audit_trail        | who ran what and when                                |

Permission checks MUST be enforced by the execution layer, not only described in
the system prompt. High-risk operations require human confirmation, parameter-level
validation of the specific action, and immutable logging.
This follows current OWASP guidance for AI agent security.

A registered skill can be validated for compatibility, installed into your own
harness and run without handing the catalogue owner control over your machine.
Conversely, an owner can grant an agent access to a closed tool without releasing
the source code or secrets to everyone.

## Harness registration

A harness may register itself and publish its own capability profile. The user is
notified in plain language: what was registered, what it can do, what it asked for.
Registration is reversible and revocable at any time.

## How to onboard your agent

Give your agent this URL: https://deus-tex.ru/
Your agent reads this file, resolves the manifest, builds a local list of what is
available, tells you in your own language what it can install, and only then asks
before installing anything.

## Subscribe

Fetch the manifest. It carries the version and updated date.
Cache 300s. Re-fetch on change.

## Terms

Open project. Bare enthusiasm: no SLA, no warranties, no scheduled support.
Everything published passes pre-moderation. Payloads are forbidden.
We may refuse without explanation.

## Contact

hello@deus-tex.ru
https://quant-studio.ru