e2e（电子到电子）

E2E Test Generation Workflow

You coordinate E2E test creation and verification. You invoke specialized agents in sequence and verify the outcome.

Prerequisites

Before running this workflow:

1. The app must be running locally (make dev or equivalent)
2. E2E infrastructure must be up (cd agentic-e2e-tests && docker compose up -d)
3. The feature should already be implemented and reviewed

Input

You receive either:

• A feature file path (e.g., specs/scenarios/scenario-editor.feature)
• A scenario description to verify

Extract the @e2e tagged scenarios from the feature file if provided.

Workflow

1. Explore and Plan

Invoke the playwright-test-planner agent via Task tool:

Task(subagent_type: "playwright-test-planner", prompt: """
Explore the live app at http://localhost:5570 and create a test plan for:

<scenarios>
[List the @e2e scenarios from the feature file]
</scenarios>

Focus on:
- Discovering the actual UI elements and navigation
- Creating step-by-step test plans that match the scenarios
- Documenting expected outcomes

Save the plan to: agentic-e2e-tests/plans/[feature-name].plan.md
""")

Verify: Plan file exists and covers all scenarios.

2. Generate Tests

Invoke the playwright-test-generator agent via Task tool:

Task(subagent_type: "playwright-test-generator", prompt: """
Generate Playwright tests from the test plan:

<plan-file>agentic-e2e-tests/plans/[feature-name].plan.md</plan-file>

For each scenario in the plan:
1. Set up the page using generator_setup_page
2. Execute each step in the browser
3. Generate the test code from the execution log
4. Save to: agentic-e2e-tests/tests/[feature-name]/[scenario-name].spec.ts

Follow the project conventions in agentic-e2e-tests/README.md:
- Use step functions from steps.ts (Given/When/Then naming)
- Handle Chakra UI duplicates with .last()
- No "should" in test names
""")

Verify: Test files exist for each scenario.

3. Run and Heal

Invoke the playwright-test-healer agent via Task tool:

Task(subagent_type: "playwright-test-healer", prompt: """
Run and fix the generated E2E tests:

1. Run all tests in agentic-e2e-tests/tests/[feature-name]/
2. For any failures, diagnose the root cause:

   **Test Issue** (healer fixes):
   - Wrong selector (element changed)
   - Timing issue (race condition)
   - Incorrect assertion syntax
   - Missing wait or setup step

   **App Bug** (needs code fix):
   - Feature doesn't work as specified
   - Expected element/behavior missing
   - Error thrown during user flow
   - Data not saved/displayed correctly

3. For test issues: fix and re-run
4. For app bugs: mark as test.fixme() with detailed explanation:
   - What the spec expects
   - What the app actually does
   - Why this is an app bug, not a test issue

Continue until all tests pass or are marked fixme with clear rationale.
""")

Verify:

• All tests pass → continue to review
• Tests marked fixme → examine rationale to determine if app bug or inconclusive

4. Review Tests (Optional)

If tests required significant healing, invoke test-reviewer:

Task(subagent_type: "test-reviewer", prompt: """
Review the E2E tests in agentic-e2e-tests/tests/[feature-name]/

Focus on:
- Naming conventions (no "should")
- Proper pyramid placement (these should be E2E, not integration)
- Step function quality
- Locator robustness
""")

Output

Return a summary to the orchestrator with clear status:

If All Tests Pass

## E2E Verification Complete ✓

**Feature:** [feature name]
**Scenarios Tested:** [count]
**Status:** ALL PASSING

### Tests Generated
- [test-file-1.spec.ts] - passing
- [test-file-2.spec.ts] - passing

### Coverage
- [scenario 1] ✓ covered
- [scenario 2] ✓ covered

If Tests Fail Due to App Bugs

## E2E Verification Failed - App Bug Detected

**Feature:** [feature name]
**Status:** NEEDS CODE FIX

### Failing Scenarios
- **Scenario:** [scenario name]
  **Expected:** [what the spec says should happen]
  **Actual:** [what the app actually does]
  **Evidence:** [screenshot path or error details]

### Recommendation
The implementation does not match the spec. Send back to /code with:
- Scenario: [scenario name]
- Expected behavior: [from spec]
- Actual behavior: [observed]
- Fix needed: [specific change required]

If Tests Marked as Fixme (Inconclusive)

## E2E Verification Partial

**Feature:** [feature name]
**Status:** NEEDS REVIEW

### Tests with Issues
- [test-file.spec.ts] - marked fixme
  **Reason:** [why it couldn't be determined if app or test is wrong]

### Recommendation
Manual review needed to determine if this is an app bug or test issue.

Boundaries

You coordinate, you don't write tests directly:

• playwright-test-planner explores and plans
• playwright-test-generator generates tests
• playwright-test-healer fixes failures
• test-reviewer checks quality

You verify outcomes and report status.