# Unity Agent Team v2: Adaptive Multi-Agent Development Framework for Unity DOTS

> Unity Agent Team v2 is an adaptive multi-agent AI framework designed for Unity DOTS. It addresses issues in version 1 such as fixed team structures, premature parallel execution, and weak constraint mechanisms through task classification, dynamic team formation, and strict phase gating.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-05-26T03:45:24.000Z
- 最近活动: 2026-05-26T03:53:52.658Z
- 热度: 152.9
- 关键词: Unity DOTS, 多智能体, AI开发框架, ECS, 自适应流水线, 智能体编排, 游戏开发, MCP, 软件开发自动化
- 页面链接: https://www.zingnex.cn/en/forum/thread/unity-agent-team-v2-unity-dots
- Canonical: https://www.zingnex.cn/forum/thread/unity-agent-team-v2-unity-dots
- Markdown 来源: floors_fallback

---

## Unity Agent Team v2: Core Overview

Unity Agent Team v2 is an adaptive multi-agent AI framework designed for Unity DOTS (Data-Oriented Technology Stack). It addresses key limitations of v1, including fixed team structures, unenforced rules, premature parallel execution, and tmux dependency. Core improvements include task classification, dynamic team formation, stage gating, adaptive pipelines, and a skill pack system to enhance flexibility and control.

## Background: Limitations of Unity Agent Team v1

v1 had several critical flaws: fixed 4-agent structure (architect, Unity dev, data tool, tester) leading to resource waste; rules as unenforced Markdown docs; premature parallel execution causing conflict edits; tmux as a hard dependency; and nested sub-agents increasing architectural complexity.

## Adaptive Pipeline Architecture

v2's pipeline consists of 5 steps: 
1. Bootstrap: Run `orchestrate.py preflight` for environment checks and state reset.
2. Triage: Generate `triage.json` (complexity, domain, confidence) via CRG and fingerprint analysis.
3. Plan: Create `pipeline.json` (phases, parallel strategy, product list) from triage data.
4. Execute Phases: Gate check → generate agents → write products → verify/retry (max 2 times).
5. Finalize: Generate report or return exit 4 for validation failure.

## Complexity-Driven Pipeline Configuration

v2 adjusts pipeline based on task complexity:
- Tiny: Unity dev + deterministic bundle
- Small: Unity dev + verifier
- Medium: Architect + Unity dev + verifier
- Large/Critical: Full team + tester
Intent modifiers (bug/refactor/explore) override defaults; depth modifiers (quick/normal/deep) adjust complexity level (e.g., quick reduces complexity, deep increases it).

## Runtime Enforcement & Skill Pack System

Enforcement: Exit codes ensure rule compliance (0=success,2=gate violation,3=ownership issue,4=validation failure,10=retry limit). Schema validation for products (stored in `.claude/schemas/`). Skill packs replace nested agents: e.g., burst-safety for Unity devs, ownership-partitioning for parallel writers.

## Key Products & Schema Validation

Key products:
- triage.json (owner: triage)
- pipeline.json (owner: orchestrate.py plan)
- root_cause.json (owner: bug-investigation/refactor-agent)
- approved_plan.json (owner: architect for medium+ tasks)
- impl_result.json (owner: Unity dev/data tool)
- verification_result.json (owner: verifier/tester)
- ownership.lock.json (owner: architect/triage)
Each product is validated against its schema before next phase.

## Conclusion & Practical Significance

v2 balances flexibility and control in multi-agent systems. For Unity DOTS devs: it provides DOTS-specific skills (ECS/Jobs/Burst) missing in general AI tools. For AI researchers: it offers a reference for task classification, dynamic planning, and stage gating in multi-agent collaboration. It's an important evolution of AI-assisted software development frameworks.