When I Entrusted TDD to an AI Agent

说明与用途

作者使用LLM实现了自主的TDD循环，保持了对测试和架构的控制，同时减少了人工干预和信息噪声。

问题

Relying on AI agents for TDD risked losing control over test-driven development flow and architectural integrity.

结果

The autonomous TDD skill reduced author friction, maintained test‑driven flow, and enforced architectural boundaries through checkpoints and quiet output, preventing implementation‑detail loops and fragile test suites.

工作流

• 1. Author approves next behavior.
• 2. Agent writes failing test and explains expected failure.
• 3. Agent implements minimal production code to pass test.
• 4. Agent evaluates refactoring opportunity and runs relevant tests.
• 5. Agent pauses for author approval before next cycle (or continues autonomously).
• 6. Agent performs vertical checkpoints every two cycles to ensure progress on public feature path.
• 7. Agent emits only meaningful events in quiet mode (unexpected failures, architectural decisions, significant refactorings, final summary).

证据

The agent would: 1) write a comment describing the intention of the smallest test it wanted to implement; 2) wait for my approval; 3) implement the test and explain how it expected the test to fail; 4) wait for approval again; 5) implement the minimum production code necessary to make the test pass; 6) wait before refactoring or moving to the next behavior.

来源

I created another variation called tdd‑strict‑autonomous. This version no longer stopped after every completed cycle. It continued automatically through red, green, and refactor, pausing only when it encountered an exceptional situation such as ambiguous behavior, an unexpected failure, an architectural decision, or an environment problem.

来源

The agent still produced a large volume of messages. Detailed output had initially helped me trust the process. Once the agent became autonomous, however, the same volume made it harder to notice the information that actually mattered.

来源