多Agent协作平台任务分配与技能复用架构设计

You are a senior platform architect designing a multi-agent collaboration platform that turns AI coding agents into managed teammates.

Requirements

Design a system where:

1. Multiple AI coding agents (Claude Code, Codex, Cursor, etc.) work as a coordinated team
2. A human manager assigns high-level goals; the platform decomposes and distributes work
3. Agents share learned skills and context across sessions

Please Design:

1. Task Decomposition Engine

• Input: A PRD or GitHub issue
• Output: Dependency-aware task graph with estimated complexity, required skills, parallelizable vs sequential tasks
• Algorithm for optimal assignment based on agent capabilities and current load

2. Agent Skill Registry

• Schema for registering agent capabilities (languages, frameworks, tools)
• Skill scoring based on past task completion quality
• Automatic skill discovery from agent work history
• Cross-agent skill sharing protocol

3. Progress Tracking and Coordination

• Real-time task status dashboard data model
• Conflict detection when agents modify overlapping files
• Merge strategy for parallel agent outputs
• Escalation rules (when to involve human)

4. Communication Protocol

• Inter-agent message format for handoffs
• Context compression for passing relevant state between agents
• Shared memory/knowledge base architecture

5. Quality Assurance

• Automated code review between agents
• Integration test orchestration
• Regression detection across agent commits

Provide: System architecture diagram (Mermaid), API schemas, data models (TypeScript interfaces), and an example workflow.