Back to Details

build

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese
Arguments: $FEATURE
Read CLAUDE.md before doing anything else. Read the acceptance criteria for "$FEATURE" in docs/PRD.md.
━━━ STEP 1: SPEC AGENT (isolated subagent) ━━━
Spawn a subagent with ONLY this context:
  • The acceptance criteria for $FEATURE from PRD.md
  • CLAUDE.md (for naming conventions and patterns)
  • The relevant wireframe from wireframes/ if a UI feature
Subagent instruction: "Write failing tests for $FEATURE to: tests/unit/$FEATURE.test.ts tests/integration/$FEATURE.integration.test.ts
Tests must be RED. Do not write any implementation code. Write tests that would pass if the feature worked exactly as described in the acceptance criteria — not tests that are easy to make pass."
━━━ STEP 2: CONFIRM RED ━━━
Run: npm run test -- $FEATURE
If tests pass (shouldn't happen yet): something is wrong. Inspect and fix. If tests fail with import errors only: fix imports, this is not real RED. If tests fail because the feature doesn't exist: this is correct RED. Proceed.
━━━ STEP 3: IMPLEMENT AGENT (fresh subagent, isolated context) ━━━
Spawn a new subagent with ONLY this context:
  • The failing test files
  • CLAUDE.md
  • Relevant wireframe
Subagent instruction: "The test files contain failing tests for $FEATURE. Implement the minimum code necessary to make ALL tests pass. Read CLAUDE.md for patterns and conventions. Do not add any functionality that is not tested."
━━━ STEP 4: CONFIRM GREEN ━━━
Run: npm run test -- $FEATURE
If GREEN: proceed to refactor. If RED: spawn FIX AGENT: "Tests are still failing. Error output: [paste exact output]. Fix the implementation only. Do not change test assertions. The tests are correct. The implementation is wrong." Re-run. If still RED after 3 attempts: stop and report the issue.
━━━ STEP 5: REFACTOR ━━━
Spawn REFACTOR AGENT: "All tests are GREEN for $FEATURE. Refactor the implementation for clarity, performance, and consistency with CLAUDE.md patterns. Tests must remain GREEN after every change. Run tests after each significant change."
━━━ STEP 6: HAND OFF ━━━
For web applications (UI features), run BOTH: /test-ui $FEATURE — interactive browser testing via Playwright MCP /qa-run $FEATURE — full 8-agent QA loop
/test-ui explores the live app and generates initial Playwright test files. /qa-run then runs the complete QA pipeline including those generated tests.
For non-UI features (API, data layer, utilities): /qa-run $FEATURE — skip /test-ui, go straight to QA loop
参数：$FEATURE
在执行任何操作前，请先阅读CLAUDE.md。 阅读docs/PRD.md中关于"$FEATURE"的验收标准。
━━━ 步骤1：SPEC AGENT（独立子代理）━━━
生成仅包含以下上下文的子代理：
  • 来自PRD.md的$FEATURE验收标准
  • CLAUDE.md（用于命名规范和模式）
  • 若为UI特性，需包含wireframes/中的相关线框图
子代理指令： "为$FEATURE编写失败的测试，保存至： tests/unit/$FEATURE.test.ts tests/integration/$FEATURE.integration.test.ts
测试必须处于RED状态。请勿编写任何实现代码。 编写的测试应在功能完全符合验收标准描述时通过——而非编写易于通过的测试。"
━━━ 步骤2：确认RED状态 ━━━
运行命令：npm run test -- $FEATURE
若测试通过（此时不应发生）：存在问题，请检查并修复。 若测试仅因导入错误失败：修复导入问题，这不属于真正的RED状态。 若测试因特性未实现而失败：这是正确的RED状态，继续执行。
━━━ 步骤3：IMPLEMENT AGENT（全新独立子代理）━━━
生成仅包含以下上下文的新子代理：
  • 失败的测试文件
  • CLAUDE.md
  • 相关线框图
子代理指令： "测试文件中包含$FEATURE的失败测试。 编写最少必要代码以让所有测试通过。 阅读CLAUDE.md以遵循模式和规范。 请勿添加任何未被测试覆盖的功能。"
━━━ 步骤4：确认GREEN状态 ━━━
运行命令：npm run test -- $FEATURE
若为GREEN状态：进入重构步骤。 若为RED状态：生成FIX AGENT： "测试仍然失败。错误输出：[粘贴准确输出]。 仅修复实现代码，请勿修改测试断言。 测试是正确的，问题出在实现上。" 重新运行。若3次尝试后仍为RED状态：停止并报告问题。
━━━ 步骤5：重构 ━━━
生成REFACTOR AGENT： "$FEATURE的所有测试均为GREEN状态。 对实现代码进行重构，提升可读性、性能，并与CLAUDE.md中的模式保持一致。 每次修改后必须保持测试为GREEN状态。 每次重大修改后运行测试。"
━━━ 步骤6：交接 ━━━
对于Web应用（UI特性），同时运行以下两个命令： /test-ui $FEATURE —— 通过Playwright MCP进行交互式浏览器测试 /qa-run $FEATURE —— 完整的8代理QA循环
/test-ui会探索实时应用并生成初始Playwright测试文件。 /qa-run随后会运行完整的QA流水线，包括这些生成的测试。
对于非UI特性（API、数据层、工具类）： /qa-run $FEATURE —— 跳过/test-ui，直接进入QA循环