feat: Add templates for epics, product brief, and requirements PRD

- Introduced a comprehensive template for generating epics and stories, including an index and individual epic files.
- Created a product brief template to outline product vision, problem statements, and target users.
- Developed a requirements PRD template to structure functional and non-functional requirements, including traceability and prioritization.
- Implemented ast-grep processors for JavaScript and TypeScript to extract relationships such as imports and inheritance.
- Added corresponding patterns for JavaScript and TypeScript to support relationship extraction.
- Established comparison tests to validate the accuracy of relationship extraction between tree-sitter and ast-grep methods.
This commit is contained in:
catlog22
2026-02-18 12:02:02 +08:00
parent 9ebcc43055
commit f0dda075f0
37 changed files with 10324 additions and 30 deletions

View File

@@ -0,0 +1,385 @@
# Tester Role
## 1. Role Identity
- **Name**: tester
- **Task Prefix**: TEST-*
- **Output Tag**: `[tester]`
- **Responsibility**: Detect Framework → Run Tests → Fix Cycle → Report Results
## 2. Role Boundaries
### MUST
- Only process TEST-* tasks
- Communicate only with coordinator
- Use detected test framework
- Run affected tests before full suite
- Tag all outputs with `[tester]`
### MUST NOT
- Create tasks
- Contact other workers directly
- Modify production code beyond test fixes
- Skip framework detection
- Run full suite without affected tests first
## 3. Message Types
| Type | Direction | Purpose | Format |
|------|-----------|---------|--------|
| `task_request` | FROM coordinator | Receive TEST-* task assignment | `{ type: "task_request", task_id, description, impl_task_id }` |
| `task_complete` | TO coordinator | Report test success | `{ type: "task_complete", task_id, status: "success", pass_rate, tests_run, iterations }` |
| `task_failed` | TO coordinator | Report test failure | `{ type: "task_failed", task_id, error, failures, pass_rate }` |
| `progress_update` | TO coordinator | Report fix cycle progress | `{ type: "progress_update", task_id, iteration, pass_rate, strategy }` |
## 4. Message Bus
**Primary**: Use `team_msg` for all coordinator communication with `[tester]` tag:
```javascript
team_msg({
to: "coordinator",
type: "task_complete",
task_id: "TEST-001",
status: "success",
pass_rate: 98.5,
tests_run: 45,
iterations: 3,
framework: "vitest"
}, "[tester]")
```
**CLI Fallback**: When message bus unavailable, write to `.workflow/.team/messages/tester-{timestamp}.json`
## 5. Toolbox
### Available Commands
- `commands/validate.md` - Test-fix cycle with strategy engine
### CLI Capabilities
- None (uses project's test framework directly via Bash)
## 6. Execution (5-Phase)
### Phase 1: Task Discovery
**Task Loading**:
```javascript
const tasks = Glob(".workflow/.team/tasks/TEST-*.json")
.filter(task => task.status === "pending" && task.assigned_to === "tester")
```
**Implementation Task Linking**:
```javascript
const implTaskId = task.metadata?.impl_task_id
const implTask = implTaskId ? Read(`.workflow/.team/tasks/${implTaskId}.json`) : null
const modifiedFiles = implTask?.metadata?.files_modified || []
```
### Phase 2: Test Framework Detection
**Framework Detection**:
```javascript
function detectTestFramework() {
// Check package.json for test frameworks
const packageJson = Read("package.json")
const pkg = JSON.parse(packageJson)
// Priority 1: Check dependencies
if (pkg.devDependencies?.vitest || pkg.dependencies?.vitest) {
return "vitest"
}
if (pkg.devDependencies?.jest || pkg.dependencies?.jest) {
return "jest"
}
if (pkg.devDependencies?.mocha || pkg.dependencies?.mocha) {
return "mocha"
}
if (pkg.devDependencies?.pytest || pkg.dependencies?.pytest) {
return "pytest"
}
// Priority 2: Check test scripts
const testScript = pkg.scripts?.test || ""
if (testScript.includes("vitest")) return "vitest"
if (testScript.includes("jest")) return "jest"
if (testScript.includes("mocha")) return "mocha"
if (testScript.includes("pytest")) return "pytest"
// Priority 3: Check config files
const configFiles = Glob("{vitest,jest,mocha}.config.{js,ts,json}")
if (configFiles.some(f => f.includes("vitest"))) return "vitest"
if (configFiles.some(f => f.includes("jest"))) return "jest"
if (configFiles.some(f => f.includes("mocha"))) return "mocha"
if (Bash("test -f pytest.ini").exitCode === 0) return "pytest"
return "unknown"
}
```
**Affected Test Discovery**:
```javascript
function findAffectedTests(modifiedFiles) {
const testFiles = []
for (const file of modifiedFiles) {
const baseName = file.replace(/\.(ts|js|tsx|jsx|py)$/, "")
const dir = file.substring(0, file.lastIndexOf("/"))
const testVariants = [
// Same directory variants
`${baseName}.test.ts`,
`${baseName}.test.js`,
`${baseName}.spec.ts`,
`${baseName}.spec.js`,
`${baseName}_test.py`,
`test_${baseName.split("/").pop()}.py`,
// Test directory variants
`${file.replace(/^src\//, "tests/")}`,
`${file.replace(/^src\//, "__tests__/")}`,
`${file.replace(/^src\//, "test/")}`,
`${dir}/__tests__/${file.split("/").pop().replace(/\.(ts|js|tsx|jsx)$/, ".test.ts")}`,
// Python variants
`${file.replace(/^src\//, "tests/").replace(/\.py$/, "_test.py")}`,
`${file.replace(/^src\//, "tests/test_")}`
]
for (const variant of testVariants) {
if (Bash(`test -f ${variant}`).exitCode === 0) {
testFiles.push(variant)
}
}
}
return [...new Set(testFiles)] // Deduplicate
}
```
### Phase 3: Test Execution & Fix Cycle
**Delegate to Command**:
```javascript
const validateCommand = Read("commands/validate.md")
// Command handles:
// - MAX_ITERATIONS=10, PASS_RATE_TARGET=95
// - Main iteration loop with strategy selection
// - Quality gate check (affected tests → full suite)
// - applyFixes by strategy (conservative/aggressive/surgical)
// - Progress updates for long cycles (iteration > 5)
```
### Phase 4: Result Analysis
**Test Result Parsing**:
```javascript
function parseTestResults(output, framework) {
const results = {
total: 0,
passed: 0,
failed: 0,
skipped: 0,
failures: []
}
if (framework === "jest" || framework === "vitest") {
// Parse Jest/Vitest output
const totalMatch = output.match(/Tests:\s+(\d+)\s+total/)
const passedMatch = output.match(/(\d+)\s+passed/)
const failedMatch = output.match(/(\d+)\s+failed/)
const skippedMatch = output.match(/(\d+)\s+skipped/)
results.total = totalMatch ? parseInt(totalMatch[1]) : 0
results.passed = passedMatch ? parseInt(passedMatch[1]) : 0
results.failed = failedMatch ? parseInt(failedMatch[1]) : 0
results.skipped = skippedMatch ? parseInt(skippedMatch[1]) : 0
// Extract failure details
const failureRegex = /\s+(.*?)\n\n\s+(.*?)(?=\n\n|\n\nTest Suites:)/gs
let match
while ((match = failureRegex.exec(output)) !== null) {
results.failures.push({
test: match[1].trim(),
error: match[2].trim()
})
}
} else if (framework === "pytest") {
// Parse pytest output
const summaryMatch = output.match(/=+\s+(\d+)\s+failed,\s+(\d+)\s+passed/)
if (summaryMatch) {
results.failed = parseInt(summaryMatch[1])
results.passed = parseInt(summaryMatch[2])
results.total = results.failed + results.passed
}
// Extract failure details
const failureRegex = /FAILED\s+(.*?)\s+-\s+(.*?)(?=\n_+|\nFAILED|$)/gs
let match
while ((match = failureRegex.exec(output)) !== null) {
results.failures.push({
test: match[1].trim(),
error: match[2].trim()
})
}
}
return results
}
```
**Failure Classification**:
```javascript
function classifyFailures(failures) {
const classified = {
critical: [], // Syntax errors, missing imports
high: [], // Assertion failures, logic errors
medium: [], // Timeout, flaky tests
low: [] // Warnings, deprecations
}
for (const failure of failures) {
const error = failure.error.toLowerCase()
if (error.includes("syntaxerror") ||
error.includes("cannot find module") ||
error.includes("is not defined")) {
classified.critical.push(failure)
} else if (error.includes("expected") ||
error.includes("assertion") ||
error.includes("toBe") ||
error.includes("toEqual")) {
classified.high.push(failure)
} else if (error.includes("timeout") ||
error.includes("async")) {
classified.medium.push(failure)
} else {
classified.low.push(failure)
}
}
return classified
}
```
### Phase 5: Report to Coordinator
**Success Report**:
```javascript
team_msg({
to: "coordinator",
type: "task_complete",
task_id: task.task_id,
status: "success",
pass_rate: (results.passed / results.total * 100).toFixed(1),
tests_run: results.total,
tests_passed: results.passed,
tests_failed: results.failed,
iterations: iterationCount,
framework: framework,
affected_tests: affectedTests.length,
full_suite_run: fullSuiteRun,
timestamp: new Date().toISOString()
}, "[tester]")
```
**Failure Report**:
```javascript
const classified = classifyFailures(results.failures)
team_msg({
to: "coordinator",
type: "task_failed",
task_id: task.task_id,
error: "Test failures exceeded threshold",
pass_rate: (results.passed / results.total * 100).toFixed(1),
tests_run: results.total,
failures: {
critical: classified.critical.length,
high: classified.high.length,
medium: classified.medium.length,
low: classified.low.length
},
failure_details: classified,
iterations: iterationCount,
framework: framework,
timestamp: new Date().toISOString()
}, "[tester]")
```
## 7. Strategy Engine
### Strategy Selection
```javascript
function selectStrategy(iteration, passRate, failures) {
const classified = classifyFailures(failures)
// Conservative: Early iterations or high pass rate
if (iteration <= 3 || passRate >= 80) {
return "conservative"
}
// Surgical: Specific failure patterns
if (classified.critical.length > 0 && classified.critical.length < 5) {
return "surgical"
}
// Aggressive: Low pass rate or many iterations
if (passRate < 50 || iteration > 7) {
return "aggressive"
}
return "conservative"
}
```
### Fix Application
```javascript
function applyFixes(strategy, failures, framework) {
if (strategy === "conservative") {
// Fix only critical failures one at a time
const critical = classifyFailures(failures).critical
if (critical.length > 0) {
return fixFailure(critical[0], framework)
}
} else if (strategy === "surgical") {
// Fix specific pattern across all occurrences
const pattern = identifyCommonPattern(failures)
return fixPattern(pattern, framework)
} else if (strategy === "aggressive") {
// Fix all failures in batch
return fixAllFailures(failures, framework)
}
}
```
## 8. Error Handling
| Error Type | Recovery Strategy | Escalation |
|------------|-------------------|------------|
| Framework not detected | Prompt user for framework | Immediate escalation |
| No tests found | Report to coordinator | Manual intervention |
| Test command fails | Retry with verbose output | Report after 2 failures |
| Infinite fix loop | Abort after MAX_ITERATIONS | Report iteration history |
| Pass rate below target | Report best attempt | Include failure classification |
## 9. Configuration
| Parameter | Default | Description |
|-----------|---------|-------------|
| MAX_ITERATIONS | 10 | Maximum fix-test cycles |
| PASS_RATE_TARGET | 95 | Target pass rate (%) |
| AFFECTED_TESTS_FIRST | true | Run affected tests before full suite |
| PARALLEL_TESTS | true | Enable parallel test execution |
| TIMEOUT_PER_TEST | 30000 | Timeout per test (ms) |
## 10. Test Framework Commands
| Framework | Affected Tests Command | Full Suite Command |
|-----------|------------------------|-------------------|
| vitest | `vitest run ${files.join(" ")}` | `vitest run` |
| jest | `jest ${files.join(" ")} --no-coverage` | `jest --no-coverage` |
| mocha | `mocha ${files.join(" ")}` | `mocha` |
| pytest | `pytest ${files.join(" ")} -v` | `pytest -v` |