Intermediate Stage

The core problem solved at the intermediate stage is: how to make AI not only generate usable code occasionally, but also complete engineering tasks stably and controllably.

Learning sequence

1. From Rules to AGENTS.md: Let the Agent understand the project rules.
2. From requirement description to specification-driven development: Make task boundaries clear and verifiable.
3. From code generation to verification engineering: Use testing and evidence to ensure output quality.
4. Understand MCP, Tools and capability boundaries: securely connect to external systems.
5. Use Chrome DevTools MCP to complete the closed loop of browser debugging: let the Agent reproduce, collect evidence, repair and verify by itself.

Stage closed loop

Project Rules → Requirements Specification → Agent Execution → Result Verification

Stage acceptance

• Allows Agents to work within clear project rules and modifications.
• Able to convert fuzzy requirements into executable and verifiable specifications.
• Ability to ask AI to run tests, check results and provide proof of delivery.
• Be able to determine whether a capability should use built-in tools, MCP or manual operation.
• Access browser debugging capabilities for AI Coding Agent, and use real runtime evidence to complete front-end debugging.