Project Desk Proof

Curated technical proof artifact for Project Desk, an agent-native work desk built on the Atlier platform, by Wilds, Inc.

This repository is intentionally not the production source tree. It is a public-reviewable architecture and product-thinking artifact that shows the platform shape, tenancy model, operating discipline, and reflective-layer direction without exposing the commercially valuable core implementation.

How To Read This Repo

Start here if you are evaluating Project Desk technically:

1. Read WHAT_THIS_PROVES.md for the short evaluator view.
2. Read PLATFORM_ARCHITECTURE.md for the platform shape.
3. Read SECURITY_AND_TENANCY.md for the public-safe security posture.
4. Read ADR_DIGEST.md for decision discipline.
5. Read OPERATING_MODEL.md for the agentic build loop.
6. Read PUBLIC_BOUNDARY.md before asking why production internals are not here.

What Project Desk Demonstrates

Project Desk is a hosted, multi-tenant work desk for people directing AI coding agents across real projects:

• a hosted MCP connector (OAuth 2.1, dynamic client registration) that any MCP-capable AI client can connect to
• a command/event architecture with one enforced write surface, so a human UI and a connected AI agent are never looking at two different truths
• per-account tenant isolation at the platform layer
• live production billing with a disciplined review process for money-path code
• cross-vendor protocol compatibility work — diagnosing and fixing real interoperability defects across independently-built AI clients
• Azure-first deployment
• AI-assisted workflow acceleration where it helps the operator, never in place of human product judgment

This proof repo deliberately avoids deep detail about the command/event storage internals, token and session mechanics, billing webhook handling, and company strategy. Those stay private.

Platform Map

flowchart TB
    Core["Atlier platform<br/>accounts, tenancy, command layer"]
    Desk["Project Desk<br/>queues, Shape, handoffs"]
    MCP["Hosted MCP connector<br/>OAuth 2.1, per-account scoping"]
    Clients["Connected AI clients<br/>Claude Code, Desktop, Codex, ChatGPT"]

    Core --> Desk
    Desk --> MCP
    Clients --> MCP

Loading

Proof Contents

File	Purpose
WHAT_THIS_PROVES.md	Short evaluator guide: what the repo is evidence for
PLATFORM_ARCHITECTURE.md	Platform shape and product relationship
ADR_DIGEST.md	Selected platform-level decisions, sanitized
OPERATING_MODEL.md	How Wilds builds and hardens software with agents
SECURITY_AND_TENANCY.md	Account isolation and auth posture at a safe level
REFLECTIVE_LAYER.md	Strategic direction for agent-native, reflective software without engine details
CODE_EXCERPTS.md	Small, sanitized platform-level TypeScript excerpts
PUBLIC_BOUNDARY.md	What this proof repo intentionally excludes
PROOF_REPO_STANDARD.md	Reusable standard for Wilds public proof repos

Evaluator Summary

This repo is evidence of:

• a real, live, hosted multi-tenant product (not a prototype)
• in-house OAuth 2.1 authorization-server implementation
• command/event architecture discipline on a production system
• adversarial-review discipline specifically for money and auth code
• cross-vendor MCP protocol interoperability engineering
• agentic engineering as a repeatable operating model
• disciplined public/private boundaries

What Is Deliberately Omitted

This proof repo does not include:

• production source code
• endpoint maps sufficient to clone the platform
• database schema or storage internals
• token, session, and billing-webhook implementation detail
• deployment workflows or real resource names beyond the public product URL
• secrets, environment names, or private URLs
• real customer, account, or transaction data
• company strategy, roadmap, or go-to-market detail

Status

Curated proof artifact. Production Project Desk / Atlier remains private and live at https://atlier.ai.