Write test cases and automation scripts with AI
Skills used in this guide (3)
写新功能、修 bug 或重构时使用：强制测试驱动开发，单元、集成与端到端测试合计覆盖率 80% 以上。
npx skills add affaan-m/ECC -gPlaywright 端到端测试模式：Page Object Model、配置、CI/CD 接入、产物管理与 flaky 测试治理。
npx skills add affaan-m/ECC -g测试驱动开发。当用户想测试优先地构建功能或修 bug、提到红-绿-重构，或想要集成测试时使用。
npx skills add mattpocock/skills -g
The biggest value of testing isn't "high coverage" — it's "catching the paths that actually break". The point of AI-generated tests: give it the implemented code or a clear behavior contract, then have it add edge cases. Don't let it invent the thing under test; a test that passes on a fake object proves nothing.
Who this is for / not for
For:
- Need unit tests / edge cases for existing functions / interfaces
- Want TDD but don't know which test to write first
- Want end-to-end (E2E) tests for key user paths
Not for:
- Code whose logic isn't decided yet (define behavior first)
- Asking AI to decide "what's correct" (you define correctness)
- Treating coverage as a KPI (green ≠ correct)
Before you start
- Code under test: the function / module / interface definition.
- Behavior contract: input/output boundaries, error branches, expectations.
- Framework: unit (Vitest/Jest/pytest), E2E (Playwright).
Steps
1. Install the skill
tdd-workflow is a test-driven workflow skill. Copy its folder into your client's skills directory (same as any Agent Skill):
| Client | Directory |
|---|---|
| Claude Code | ~/.claude/skills/ |
| Codex | ~/.codex/skills/ |
| Cursor | ~/.cursor/skills/ |
| Trae | ~/.trae/skills/ |
| CodeBuddy | ~/.codebuddy/skills/ |
| Tongyi Lingma | ~/.lingma/skills/ |
2. Give behavior, then let AI write tests
From this function, write Vitest tests:
- cover: normal input, empty, out-of-range, wrong type (one each)
- one assertion per test, name states the expectation
- don't change the implementation, only add the test file
List the 3 riskiest edges first, then write.3. TDD: red before green
Using tdd-workflow: first write a failing test (red) for "cart checkout",
describing the expected behavior; after I implement, make it pass (green),
then propose one safe refactor.4. Add E2E for key paths
Use e2e-testing for Playwright on the 2-3 highest-cost paths, not clicking everything:
Write one Playwright E2E: login through successful checkout.
Assert only key states (success toast, order id generated), not every element.Pitfalls
| Pit | Symptom | Fix |
|---|---|---|
| Fake object | AI invents nonexistent fn | Give real code, don't let it invent |
| Only happy path | Edges missed | Force empty/range/error coverage |
| Too many asserts | Don't know what failed | One behavior per test |
| Green = correct | Asserted wrong way | You define "correct" |
| E2E too wide | Brittle, slow | Only key paths |
FAQ
No framework yet?
Yes — have AI use your project's; if none, pick one first.
Trust AI tests as-is?
Run once to confirm they actually failed (red), then go green; avoid "fake green".
Must test first in TDD?
Recommended. Red-green lets behavior be defined before implementation.
How much E2E?
Only the highest-cost paths; clicking all is brittle and slow.
Low coverage?
Add edges and error branches first; more useful than normal-path padding.
Related guides
Continue across these scenes:
相关源码
- affaan-m/ECC · tdd-workflow source
- affaan-m/ECC · e2e-testing source
- mattpocock/skills · tdd source
- Agent Skill open format spec
一句话结论
结论：测试的价值在"抓住真会出事的路径"——给 AI 真实代码或行为约定，让它补边界用例，别让它凭空造被测对象。
References
Related Scenes
Guard code quality with AI (translated)
AI writes fast, but the quality gate matters more. Community consensus in three: reverse TDD (AI writes tests, then minimal impl), zero-warning gate (`--max-warnings=0`, warnings silently become the new baseline), review as a gate (every change through the same gate). Make AI output verifiable and g
Get an AI code review worth reading
The classic failure mode is twenty "suggestions" where eighteen are style preferences. A useful review contains only two kinds of information: what will break, and what will confuse the next reader. So the job is not to make AI say more. It is to forbid everything else.