DuanYun is an AI Agent OS that transforms a single-sentence idea into a fully rehearsed product specification, encompassing planning, execution, and review. It targets users needing to rapidly validate product concepts, cutting validation time from days/weeks to minutes. The system offers full, real-time observability and exportability of the entire process, enabling quick iteration and informed building decisions.

How It Works

The core approach uses a multi-stage rehearsal pipeline: Smart Clarification, Route Planning, SPEC Tree decomposition, Spec Document generation (streaming requirements, design, tasks), and Effect Preview. This is powered by a "FSD Fleet" of seven specialized AI roles (Planner, Clarifier, Researcher, Generator, Operator, Reviewer, Auditor) collaborating via capability nodes and a Docker sandbox. The entire workflow is observable in real-time via a 3D office visualization and streaming workbench, ensuring transparency and an evidence trail.

Quick Start & Requirements

• Install: Clone (git clone https://github.com/opencroc/cube-pets-office.git), cd cube-pets-office, pnpm install, then pnpm run dev:all (full stack) or pnpm run dev:frontend (browser-only).
• Prerequisites: Node.js 22+, pnpm. Docker optional for full executor mode.
• Demo: Live demo at https://opencroc.github.io/cube-pets-office/.

Highlighted Details

• Full Observability: Real-time visibility into roles, capabilities, ReAct cycles, and artifacts, eliminating "black boxes."
• Multi-Route Planning: Offers distinct planning routes (Quick, Standard, Deep, Conservative) with risk/cost assessments.
• Human Takeover Governance: Integrates explicit pause points for clarification, approval, risk, budget, and delivery decisions.
• Evidence & Replay: Generates exportable artifacts, audit logs, and replay timelines for detailed decision inspection.
• Docker Sandbox: Enables secure, isolated code execution within Docker containers with live terminal streaming.

Maintenance & Community

Contribution guidelines are in CONTRIBUTING.md, detailing Git and pnpm workflows. No specific community channels are mentioned.

Licensing & Compatibility

Released under the permissive MIT License, allowing commercial use, modification, and distribution, including within closed-source applications.

Limitations & Caveats

The README does not explicitly list limitations. However, comparison highlights multi-route planning and human takeover governance as key differentiators, suggesting more robust solutions compared to alternatives. The project appears actively developed with a substantial codebase.