AI Agent 自主项目交付闭环设计器

You are an expert systems architect specializing in autonomous AI coding agent orchestration. Design a complete autonomous delivery loop for an AI coding agent team.

Context

I want to set up a system where AI coding agents can autonomously:

1. Pick up tasks from a project board (Linear/Jira/GitHub Issues)
2. Plan implementation approach
3. Write code in isolated environments
4. Run tests and CI
5. Submit PRs with proof of work
6. Handle review feedback
7. Land PRs when approved

Requirements

• Task intake: How should tasks be structured for agent consumption?
• Isolation: Each task runs in its own sandbox/branch
• Verification: Agents must provide evidence of correctness (CI green, test coverage, complexity analysis)
• Safety: What guardrails prevent agents from shipping broken code?
• Escalation: When should agents escalate to humans?
• Observability: How do we monitor agent performance and quality?

Output Format

Provide:

1. Architecture diagram (Mermaid)
2. Task specification template
3. Agent workflow state machine
4. Safety guardrail checklist
5. Metrics and KPIs to track
6. Example configuration for a TypeScript monorepo

Be specific and actionable. Include code snippets where relevant.