pcliangx/AppGenesisForge

AI Agent Team scaffold for Claude Code — 19 roles collaborate through stage gates, enforced by mandatory skills + blocking hooks + DoD checklists. Spec-driven change folders, SIT-audited code reviews, 4-layer hook defense, multi-LLM (DeepSeek/Doubao/Qwen/MiniMax). Tracks: Web · WeChat Mini Program · Apple native (macOS/iOS). 中文/EN README.

What it solves

AppGenesisForge (AGF) addresses the problem of single AI agents losing control during long, complex development cycles. Instead of relying on a single "smarter" model, AGF implements a structured, lean production line approach to software development, ensuring that defects do not flow into subsequent stages of the process.

How it works

AGF acts as a scaffolding for AI agent teams based on Claude Code Agent Teams. It organizes 19 specialized AI roles (such as Product Lead, Frontend/Backend Dev, Reviewer, and QA) into a strict 7-stage pipeline:

  1. Requirement Clarification (via change folders/OpenSpec)
  2. Task Assignment (AC extraction and worktree isolation)
  3. TDD Implementation (Unit and SIT tests first)
  4. Code Review (with SIT evidence auditing)
  5. UAT Deployment (isolated stacks and smoke tests)
  6. E2E Testing (browser control traversal)
  7. UAT Sign-off (user-verified test cases)

It enforces quality through three mechanisms: mandatory skills, hard-blocking security hooks, and Definition of Done (DoD) checklists. It also includes a real-time HTML dashboard to track task progress across these stages.

Who it’s for

Developers building Web full-stack applications (defaulting to React + FastAPI + Postgres), AI Agents/RAG systems, WeChat Mini-Programs, or native Apple applications (macOS/iOS using Swift 6 and SwiftUI).

Highlights

  • Multi-Role Orchestration: 19 distinct AI roles with locked responsibilities to prevent conflicts of interest (e.g., Reviewers cannot modify source code).
  • TDD Enforcement: Mandatory "red $\rightarrow$ green $\rightarrow$ refactor" cycle; skipping tests is grounds for review rejection.
  • Security Guardrails: Four layers of security hooks to block dangerous commands (e.g., rm -rf) and scan for leaked secrets.
  • Cross-Platform Support: Dedicated tracks for Web, WeChat Mini-Programs, and Apple native apps with specialized roles for each.
  • Real-time Monitoring: A zero-dependency HTML board that updates every 3 seconds to visualize the team's progress.
  • Design Discipline: A layer to prevent "AI-style" aesthetics by banning specific UI patterns (e.g., generic purple gradients) and enforcing design read-outs.
  • Multi-LLM Compatibility: Support for various models including DeepSeek, Doubao, Qwen, and MiniMax.