maestro-flow 0.5.2 → 0.5.31

package/workflows/auto-test.md

@@ -1,699 +1,699 @@
-# Auto-Test Workflow (Unified)
-
-Unified automated testing with intelligent state-based routing. Merges test generation (gap-forward), business testing (PRD-forward), and integration testing (code-forward) into a single pipeline.
-
-**Core idea: read project state → auto-select scenario source → shared pipeline from plan through iteration.**
-
-Pipeline: Parse → Route → Source Scenarios → Discover Infrastructure → Plan → Write Tests → Execute → Iterate → Report
-Only Step 2 diverges (scenario source). Everything else is shared.
-
----
-
-### Step 0: Parse Input & Load Specs
-
-**Parse arguments:**
-
-| Input | Result |
-|-------|--------|
-| No arguments | Error E001 |
-| Phase number | Resolve phase dir from artifact registry |
-| `--max-iter N` | Set MAX_ITER = N (default 5). **1 = single-pass** (no iteration loop) |
-| `--layer L` | Start from / restrict to specific layer (L0\|L1\|L2\|L3) |
-| `--dry-run` | Generate plan only, do not execute |
-| `--re-run` | Re-run only previously failed scenarios |
-
-**Resolve phase dir:** from `state.json` artifact registry (`type='execute'`, matching phase). Error E002 if not found.
-
-**Load specs:**
-```
-specs_test = maestro spec load --category test
-specs_arch = maestro spec load --category arch
-```
-
-`specs_test` for test conventions (Steps 3-5); `specs_arch` for module boundaries (Step 2).
-
----
-
-### Step 1: Read State & Route
-
-Read project state signals and auto-select scenario source. This is the **primary branch point, with Route-specific extraction in Step 2** (after Step 2 normalization, the downstream pipeline is identical).
-
-```
-Priority: Resume > Re-run > Spec > Gap > Code
-
-1. RESUME:
-   Check: ${PHASE_DIR}/.tests/auto-test/state.json exists AND status == "running"
-   Compat: also check ${PHASE_DIR}/.tests/integration/state.json (old format)
-   Action: offer resume or restart
-   If resume: restore session, jump to Step 6 at saved iteration/layer
-   If restart: archive to .history/, continue as new
-
-2. RE-RUN:
-   Check: --re-run flag AND .tests/auto-test/report.json has failed/blocked scenarios
-   Compat: also check .tests/business/business-test-report.json (old format)
-   Action: load failed/blocked scenarios with status reset to pending
-   Skip to Step 4 (scenarios pre-loaded, plan confirmation)
-
-# Note: `integration/state.json` and `business/business-test-report.json` are
-# legacy compat paths read for backward compatibility only. The new pipeline
-# writes exclusively to `.tests/auto-test/state.json` and `.tests/auto-test/report.json`
-# (these legacy paths are superseded; see migration table at end of file).
-
-3. SPEC:
-   Check: .workflow/blueprint/BLP-*/requirements/REQ-*.md exists
-   Resolve: SPEC_DIR from index.json.blueprint_ref or most recent BLP-*/
-   If SPEC_DIR found: set ROUTE = "spec", SPEC_MODE = "full"
-   If no spec but has success_criteria: set ROUTE = "spec", SPEC_MODE = "degraded"
-
-4. GAP:
-   Check: ${PHASE_DIR}/verification.json has gaps[] with status MISSING or PARTIAL
-   OR: ${PHASE_DIR}/.tests/coverage-report.json has requirements_uncovered[]
-   Action: set ROUTE = "gap"
-
-5. CODE:
-   Default fallback when no spec package and no gaps detected.
-   Action: set ROUTE = "code"
-```
-
-**Display route selection:**
-```
-=== AUTO-TEST ===
-来源: {ROUTE} ({reason})
-阶段: {phase_name}
-{IF spec: "Spec: {SPEC_DIR} (mode: {SPEC_MODE})"}
-```
-
----
-
-### Step 2: Source Scenarios
-
-Execute the route-specific scenario extraction, then normalize ALL scenarios into the unified format. **After this step, the pipeline is identical regardless of source.**
-
-#### Route A: `spec` source (PRD-forward)
-
-1. Load spec package:
-   - `requirements/REQ-*.md` — functional requirements with acceptance criteria
-   - `requirements/NFR-*.md` — non-functional requirements
-   - `architecture/_index.md` — API endpoints, data model, state machines
-   - `epics/EPIC-*.md` — user stories for E2E scenarios
-
-2. For each `REQ-NNN-{slug}.md`, parse `## Acceptance Criteria`:
-   - Extract each numbered criterion as a scenario seed
-   - Map RFC 2119 keywords to priority:
-
-   | Keyword | Priority | Failure Severity |
-   |---------|----------|-----------------|
-   | MUST / MUST NOT / SHALL / SHALL NOT | critical | blocker |
-   | SHOULD / SHOULD NOT / RECOMMENDED | high | major |
-   | MAY / OPTIONAL | medium | minor |
-
-3. Classify scenario into layer:
-
-   | Source | Layer | Category |
-   |--------|-------|----------|
-   | Architecture API endpoints + REQ AC about request/response | L1 | api_contract |
-   | REQ AC about business logic, validation, state changes | L2 | business_rule |
-   | REQ AC about state transitions (from architecture state machines) | L2 | state_transition |
-   | Epic user stories (multi-step flows) | L3 | user_flow |
-   | NFR performance/security constraints | L2 | non_functional |
-
-4. **Generate Fixtures** (conditional sub-step):
-
-   **Tier 1: Schema-derived fixtures**
-   ```
-   FOR each entity in REQ data model:
-     valid: object satisfying all field constraints
-     invalid: one variant per constraint violation (null_required, empty_string, overflow, wrong_type)
-     boundary: edge value variants (min_value, max_value, min_minus_one, max_plus_one)
-   ```
-
-   **Tier 2: Acceptance-criteria-derived expectations**
-   ```
-   FOR each "MUST return X when Y": fixture { input: Y, expected: X }
-   FOR each "MUST validate Z": fixture { input: invalid_Z, expected: validation_error }
-   ```
-
-   **Tier 3: Business-scenario-derived data sets (L3 only)**
-   ```
-   FOR each Epic user story:
-     scenario_pack: coordinated data set across story steps
-   ```
-
-   **Mock contracts:**
-   ```
-   FOR each API endpoint in architecture/_index.md:
-     mock_contract: { request_pattern, response_fixture }
-   ```
-
-5. **Degraded mode** (SPEC_MODE = "degraded"):
-   - Extract from `index.json.success_criteria` (each → one L2 scenario)
-   - Extract from `plan.json` task convergence criteria (each → one L1/L2 scenario)
-   - Extract from `.summaries/TASK-*-summary.md` (each → one L1 scenario)
-   - All default to priority: "high". No L3 in degraded mode.
-
-6. Convert to unified scenario format.
-
-#### Route B: `gap` source (coverage-forward)
-
-1. Read gap sources:
-   - `verification.json` → `gaps[]` with status MISSING or PARTIAL
-   - `coverage-report.json` → `requirements_uncovered[]`
-   - Task summaries → modified files list
-
-2. Priority: MISSING or uncovered → HIGH; PARTIAL → MEDIUM.
-
-3. Classify each changed file:
-
-   | File Type | Category | Layer |
-   |-----------|----------|-------|
-   | Pure function / utility | unit | L1 |
-   | React component | unit + e2e | L1 + L3 |
-   | API route / handler | integration | L2 |
-   | Database model / query | integration | L2 |
-   | CLI command | e2e | L3 |
-   | Config / types / constants / CSS / test files | skip | — |
-
-4. **Optional: CLI supplementary analysis** (skip if no CLI tools enabled or all files "skip"):
-   ```
-   Bash({
-     command: 'maestro delegate "PURPOSE: Analyze source files for test-worthy edge cases
-   TASK: Identify error handling | boundary conditions | state transitions | external dependencies
-   MODE: analysis
-   CONTEXT: @${target_files}
-   EXPECTED: JSON array of { file, edge_cases: [{ description, type, priority }] }
-   CONSTRAINTS: Non-obvious cases only | Max 5 per file
-   " --role analyze --mode analysis',
-     run_in_background: true
-   })
-   ```
-   On callback: merge edge_cases into scenarios, mark `source: "cli-analysis"`.
-
-5. Convert to unified scenario format.
-
-#### Route C: `code` source (exploration-forward)
-
-1. Explore codebase for testable integration points:
-   - Module boundaries and cross-module calls
-   - API endpoints and their handlers
-   - Database interactions and queries
-   - External service integrations
-   - Event flows and message passing
-
-2. Scan for cross-module imports, API route definitions, database calls.
-
-3. Map integration points: which modules communicate through what interfaces.
-
-4. Infer layer from integration type:
-   - Isolated function → L1
-   - Cross-module call / API handler → L2
-   - Full user flow → L3
-
-5. Convert to unified scenario format.
-
-#### Unified Scenario Format
-
-All routes produce this identical structure:
-
-```json
-{
-  "id": "AT-{NNN}",
-  "source": "spec|gap|code|re-run",
-  "layer": "L0|L1|L2|L3",
-  "priority": "critical|high|medium",
-  "category": "api_contract|business_rule|state_transition|user_flow|non_functional|unit|integration|e2e|static",
-  "name": "descriptive scenario name",
-  "target_file": "src/path/to/file.ts",
-  "test_file": "src/path/__tests__/file.test.ts",
-  "req_ref": "REQ-NNN:AC-N | gap-id | null",
-  "description": "what this scenario validates",
-  "test_cases": ["case 1", "case 2"],
-  "input": { "$fixture_ref": "..." },
-  "expected": {
-    "status": 200,
-    "behavior": "description of expected outcome"
-  },
-  "preconditions": [],
-  "postconditions": [],
-  "mock_services": [],
-  "fixtures": {}
-}
-```
-
-**Field population by route:**
-
-| Field | spec | gap | code |
-|-------|------|-----|------|
-| `req_ref` | REQ-NNN:AC-N | gap ID from verification.json | null |
-| `fixtures` | Tier 1-3 generated | empty (inferred in Step 5) | empty (inferred in Step 5) |
-| `mock_services` | from architecture/_index.md | empty | discovered from imports |
-| `preconditions` | from REQ AC text | empty | empty |
-
----
-
-### Step 3: Discover Test Infrastructure
-
-Detect existing test framework and patterns:
-
-1. **Config files**: `jest.config.*`, `vitest.config.*`, `pytest.ini`, `pyproject.toml`, `.mocharc.*`
-2. **Existing tests**: `*.test.*`, `*.spec.*`, `test_*` (exclude node_modules, .git)
-3. **Utilities**: `test-utils.*`, `testHelper*`, `conftest.py`, `setup.*`
-
-Extract: framework, directory structure, naming convention, test utilities, run command.
-
-Read 2-3 existing test files to learn: import style, describe/it nesting, assertion library, mock patterns, setup/teardown.
-
-**Detect tech stack for generation:**
-
-| Stack | L1 | L2 | L3 |
-|-------|----|----|-----|
-| Java/Spring Boot | MockMvc | JUnit 5 + WireMock | TestContainers |
-| TypeScript/Node | vitest | supertest + nock | playwright/cypress |
-| Python | pytest | httpx + responses | selenium |
-
-If no test framework detected: Error E003.
-
-Output: `infrastructure` object passed to Steps 5-6.
-
----
-
-### Step 4: Generate Test Plan & Confirm
-
-1. **Merge pre-existing tests** from `.tests/test-gen-report.json` (if exists):
-   - Mark as "pre-existing" so Step 5 skips writing them
-   - Step 6 includes them in execution
-
-2. Apply `--layer` filter if specified.
-
-3. **Archive previous plan** to `.history/` if exists.
-
-4. Write `.tests/auto-test/test-plan.json`:
-```json
-{
-  "phase": "{phase}",
-  "source_route": "{ROUTE}",
-  "blueprint_ref": "{SPEC_DIR name or null}",
-  "spec_mode": "full|degraded|null",
-  "generated_at": "{ISO timestamp}",
-  "infrastructure": { "framework": "...", "run_command": "..." },
-  "layers": {
-    "L0": { "scenario_count": N, "commands": ["tsc --noEmit", "eslint src/"] },
-    "L1": { "scenario_count": N, "priority_distribution": { "critical": X, "high": Y, "medium": Z } },
-    "L2": { "scenario_count": N, "priority_distribution": { ... } },
-    "L3": { "scenario_count": N, "priority_distribution": { ... } }
-  },
-  "scenarios": [ "...unified format..." ],
-  "fixtures": { "...if spec source..." },
-  "requirement_coverage_plan": { "requirements_targeted": [], "requirements_skipped": [] }
-}
-```
-
-5. Display and confirm:
-```
-=== AUTO-TEST PLAN ===
-来源:  {ROUTE}
-阶段:  {phase_name}
-Blueprint: {blueprint_ref or "N/A"}
-
-层级分布:
-  L0 Static:      {N} checks
-  L1 Unit/API:    {N} scenarios ({X} critical, {Y} high)
-  L2 Integration: {N} scenarios ({X} critical, {Y} high)
-  L3 E2E:         {N} scenarios ({X} critical, {Y} high)
-
-Total: {N} scenarios, {M} test cases
-Max iterations: {max_iter}
-
-Proceed? (yes/edit/cancel)
-```
-
-- `--dry-run`: stop here, report plan
-- User "edit": modify plan interactively
-- User "cancel": abort
-
----
-
-### Step 5: Write Tests (RED-GREEN) via CSV Parallel
-
-**Parallel strategy**: Build `layer-L{N}-write.csv` per layer, execute via `spawn_agents_on_csv`. Each agent writes one test file independently.
-
-#### 5a. Build Write CSV
-
-For each layer (L1, L2, L3 — sequential):
-
-```
-Extract pending scenarios for this layer from test-plan.json
-Build layer-L{N}-write.csv:
-  Columns: id, name, target_file, test_file, description, test_cases, fixtures, req_ref, infrastructure_hints, prev_context
-
-  prev_context = findings from completed prior-layer scenarios (cross-layer propagation)
-```
-
-#### 5b. Parallel Test Writing via spawn_agents_on_csv
-
-```javascript
-spawn_agents_on_csv({
-  csv_path: `.tests/auto-test/.csv-session/layer-L${N}-write.csv`,
-  id_column: "id",
-  instruction: `
-    Write ONE test file for the given scenario using RED-GREEN methodology.
-
-    Rules:
-    - Read target_file to understand module under test
-    - Write test at test_file path following infrastructure_hints patterns
-    - Each test_case → one it() block, include id in describe("AT-NNN: {name}")
-    - Use fixtures column (infer from source if empty)
-    - Run test once after writing: report red_result
-    - NEVER modify source code — only write/fix test files
-
-    RED results:
-    - pass: test passes immediately (may need strengthening)
-    - expected_fail: test correctly targets real behavior
-    - unexpected_fail: setup/import error — fix test, re-run
-
-    Read discoveries.ndjson for shared patterns. Append if you find reusable ones.
-  `,
-  max_concurrency: 5,
-  max_runtime_seconds: 1800,
-  output_csv_path: `.tests/auto-test/.csv-session/layer-L${N}-write-results.csv`,
-  output_schema: { id, status: [written|failed], red_result: [expected_fail|unexpected_fail|pass], findings, error }
-})
-```
-
-#### 5c. Merge & Continue
-
-Merge write-results into master state. Delete temp CSV. Proceed to next layer or Step 6.
-
-**If `--max-iter 1`:** After all layers written and run once, jump directly to Step 8 (single-pass mode, replaces test-gen behavior). Skip Steps 6-7.
-
----
-
-### Step 6: Execute (Progressive Layers)
-
-Run tests progressively through layers with fail-fast on critical:
-
-- L0: `tsc --noEmit` + `eslint src/` (static analysis)
-- L1: unit tests (`--testPathPattern="unit|__tests__"`)
-- L2: integration tests (`--testPathPattern="integration"`)
-- L3: E2E tests (`--testPathPattern="e2e"`)
-
-**Fail-fast rule:**
-- L0 must pass before L1
-- If ANY "critical" priority failure in L1 → do NOT proceed to L2
-- If ANY "critical" priority failure in L2 → do NOT proceed to L3
-- "high" and "medium" failures do NOT block next layer
-
-Record per-scenario results: `{ status, actual_response, duration_ms, error_detail, classification: null }`.
-
-Write iteration results to `.tests/auto-test/results-iter-{N}.json`.
-
----
-
-### Step 7: Reflect & Adjust (Unified Iteration Engine)
-
-Single engine that subsumes both Generator-Critic (per-layer inner) and Reflect-Adjust (global outer). Uses `spawn_agents_on_csv` for parallel failure diagnosis.
-
-```
-OUTER LOOP (max_iter iterations):
-
-  FOR each active layer (L0 through current):
-
-    INNER LOOP (max 3 iterations per layer — Generator-Critic):
-
-      1. Build diagnosis CSV from failed scenarios:
-
-         Build diagnosis-iter-{N}.csv:
-           Columns: id, scenario_id, layer, test_file, error_detail, expected, actual, target_file, source_context
-
-         spawn_agents_on_csv({
-           csv_path: `.tests/auto-test/.csv-session/diagnosis-iter-${iter}.csv`,
-           id_column: "id",
-           instruction: `
-             Classify ONE test failure and provide fix if applicable.
-
-             Classifications:
-             - test_defect: Test wrong (bad import, endpoint, fixture, assertion)
-             - code_defect: Source violates business rule (actual != expected)
-             - env_issue: Environment problem (service down, config missing)
-
-             If test_defect: provide fix_code (old → new)
-             If code_defect/env_issue: leave fix_code empty, provide evidence
-
-             Rules:
-             - NEVER suggest source code changes
-             - A test correctly catching a real bug = code_defect, not test_defect
-             - When uncertain: prefer code_defect (conservative)
-           `,
-           max_concurrency: 5,
-           max_runtime_seconds: 1200,
-           output_csv_path: `.tests/auto-test/.csv-session/diagnosis-iter-${iter}-results.csv`,
-           output_schema: { id, classification, fix_code, evidence, error }
-         })
-
-      2. Merge diagnosis results:
-         - test_defect with fix_code → apply fix to test file
-         - code_defect → mark as confirmed failure (stop retrying)
-         - env_issue → mark as blocked
-
-      3. IF test_defects found:
-         Re-run ALL scenarios in this layer (catch regressions)
-
-      4. IF no test_defects remain: break inner loop
-
-    END INNER
-
-    Record final layer results
-    IF critical code_defects in this layer: stop layer progression (fail-fast)
-
-  END FOR
-
-  REFLECT:
-    Analyze: which tests failed, pass rate improving/plateauing/regressing,
-    failures clustered by component, strategy effectiveness.
-    Append to reflection-log.md:
-      iteration, strategy, pass_rate delta, what worked/failed,
-      detected patterns, strategy assessment (effective/ineffective + recommendation)
-
-  ADJUST (Adaptive Strategy Engine):
-
-    | Condition | Strategy | Behavior |
-    |-----------|----------|----------|
-    | Iteration 1-2 | Conservative | Fix obvious test_defects only, don't refactor |
-    | Pass rate >80% AND failures similar | Aggressive | Batch-fix related failures together |
-    | New regressions appeared | Surgical | Revert last changes, fix regression only |
-    | Stuck 3+ iterations (rate not improving) | Reflective | Step back, re-analyze root cause pattern |
-
-    Transitions:
-      Conservative --(>80%)--> Aggressive
-      Aggressive --(regression)--> Surgical --(fixed)--> Aggressive
-      Any --(stuck 3+ iters)--> Reflective --(insight)--> Conservative
-
-  CONVERGENCE CHECK:
-    pass_rate >= threshold (95%) → Step 8 (converged)
-    iteration >= max_iter → Step 8 (max_iter_reached)
-    all remaining failures = code_defect → Step 8 (confirmed_defects)
-    ELSE → next outer iteration (back to Execute)
-
-END OUTER
-```
-
-**Degenerate cases:**
-- `max_iter=1`: Step 5 writes tests, Step 6 executes once, Step 7 runs Reflect only (log results), no Adjust, no loop → Step 8. Equivalent to test-gen single pass.
-- `max_iter=3`: Inner loop cleans test_defects. Up to 3 outer passes. Similar to business-test.
-- `max_iter=5`: Full adaptive strategy progression. Similar to integration-test.
-
----
-
-### Step 7.5: Test Confidence Scoring
-
-Scored after each REFLECT step. Dimensions (5): scenario_coverage, test_quality, diagnostic_accuracy, strategy_effectiveness, infrastructure_fitness. Factors (weights): completeness(.30), pass_rate_trend(.25), classification_accuracy(.20), coverage_breadth(.15), consistency(.10). Append confidence table to reflection-log.md.
-
-**Enhanced Convergence**: pass_rate ≥ 95% AND confidence ≥ 60% → converged. pass_rate ≥ 95% BUT confidence < 60% → continue (tests may be weak). max_iter reached or all failures = code_defect → Step 8.
-
-**Quality mechanisms**: Pressure Pass (before Step 8) — select 2-3 passing tests from highest-pass-rate layer, verify they exercise real behavior (not mock-only, non-trivial assertions). Devil's Advocate — pass_rate > 80% → challenge assertion specificity, error path coverage, mock over-reliance. Stall Detection — delta < 5% for 2 iterations + pass_rate flat → force Reflective strategy.
-
-**Readiness Gate** (before Step 8, skip if max_iter=1): scenario_coverage < 40% | no pressure pass | diagnostic_accuracy < 40% | unclassified failures. If blocked → force one additional iteration. Add confidence section to report.json.
-
----
-
-### Step 8: Complete & Write Artifacts
-
-1. Update session state:
-```json
-// .tests/auto-test/state.json
-{
-  "session_id": "auto-test-{YYYYMMDD-HHmmss}",
-  "phase": "{phase}",
-  "phase_dir": "{PHASE_DIR}",
-  "source_route": "spec|gap|code|re-run",
-  "blueprint_ref": "SPEC-001 | null",
-  "spec_mode": "full|degraded|null",
-  "status": "converged|max_iter_reached|confirmed_defects|single_pass",
-  "flags": { "max_iter": 5, "layer": null, "dry_run": false, "re_run": false },
-  "iteration": 2,
-  "strategy": "conservative",
-  "strategy_history": ["conservative", "aggressive"],
-  "threshold": 95,
-  "current_layer": "L2",
-  "layer_state": {
-    "L0": { "inner_iter": 1, "pass_rate": 100.0, "status": "passed" },
-    "L1": { "inner_iter": 2, "pass_rate": 95.0, "status": "passed" },
-    "L2": { "inner_iter": 1, "pass_rate": 87.5, "status": "completed" },
-    "L3": { "inner_iter": 0, "pass_rate": 0, "status": "pending" }
-  },
-  "pass_rate_history": [72.0, 85.0, 95.0],
-  "scenario_count": 30,
-  "infrastructure": { "framework": "vitest", "run_command": "npm test" },
-  "started_at": "{ISO}",
-  "updated_at": "{ISO}"
-}
-```
-
-2. Archive previous report to `.history/` if exists.
-
-3. Write `.tests/auto-test/report.json`:
-```json
-{
-  "phase": "{phase}",
-  "source_route": "{ROUTE}",
-  "blueprint_ref": "{spec ref or null}",
-  "spec_mode": "full|degraded|null",
-  "completed_at": "{ISO timestamp}",
-  "convergence": {
-    "status": "converged|max_iter_reached|confirmed_defects|single_pass",
-    "iterations": N,
-    "final_pass_rate": 95.0,
-    "threshold": 95,
-    "strategy_history": ["conservative", "aggressive"]
-  },
-  "infrastructure": { "framework": "vitest", "test_dir": "__tests__/", "run_command": "npm test" },
-  "layers": {
-    "L0": { "total": N, "passed": P, "failed": F, "blocked": B, "pass_rate": 100.0 },
-    "L1": { "total": N, "passed": P, "failed": F, "blocked": B, "pass_rate": 95.0 },
-    "L2": { ... },
-    "L3": { ... }
-  },
-  "scenarios": [
-    {
-      "id": "AT-001", "source": "spec", "layer": "L1", "name": "...",
-      "test_file": "...", "req_ref": "REQ-001:AC-1",
-      "status": "passed|failed|blocked",
-      "classification": "null|test_defect|code_defect|env_issue",
-      "iterations_to_pass": 2
-    }
-  ],
-  "failures": [
-    {
-      "id": "AF-001", "scenario_id": "AT-005", "req_ref": "REQ-001:AC-3",
-      "layer": "L1", "severity": "critical", "classification": "code_defect",
-      "description": "...", "expected": "...", "actual": "...",
-      "fix_suggestion": { "file": "src/...", "line": 42, "direction": "..." }
-    }
-  ],
-  "requirement_coverage": [],
-  "summary": {
-    "source_route": "{ROUTE}",
-    "total_scenarios": 30, "total_passed": 26, "total_failed": 3, "total_blocked": 1,
-    "bugs_discovered": 3, "test_defects_fixed": 5, "coverage_pct": 85.0
-  }
-}
-```
-
-4. Update `validation.json`: if gap source, change MISSING → COVERED for passing tests.
-5. Update `index.json` with `auto_test` section.
-
-**Artifact file mapping (old → new):**
-
-| Old file | New file |
-|----------|----------|
-| `.tests/test-gen-report.json` | `.tests/auto-test/report.json` (source_route: "gap") |
-| `.tests/integration/state.json` | `.tests/auto-test/state.json` |
-| `.tests/integration/summary.json` | Merged into report.json |
-| `.tests/integration/reflection-log.md` | `.tests/auto-test/reflection-log.md` |
-| `.tests/business/business-test-report.json` | `.tests/auto-test/report.json` (source_route: "spec") |
-| `.tests/business/business-test-summary.md` | `.tests/auto-test/traceability.md` (conditional) |
-
----
-
-### Step 9: Post-Processing & Routing
-
-#### Conditional: Traceability Matrix (spec source only)
-
-Build REQ → AC → scenario → result mapping:
-
-```
-FOR each REQ in requirements targeted:
-  FOR each AC in REQ:
-    scenarios_for_ac = filter scenarios where req_ref == "REQ-NNN:AC-{N}"
-    ac_status = "passed" if ALL passed
-                "failed" if ANY failed
-                "blocked" if ANY blocked and none failed
-                "untested" if no scenarios mapped
-
-  req.coverage_pct = passed_criteria / total_criteria * 100
-  req.verdict = "verified" if all MUST+SHOULD passed
-                "partial" if some failed
-                "unverified" if all failed or untested
-```
-
-Populate `requirement_coverage[]` in report.json.
-Write `.tests/auto-test/traceability.md` (human-readable table).
-
-#### Conditional: Issue Creation (when failures exist)
-
-```
-FOR each failure in report.failures WHERE classification == "code_defect":
-  issue = {
-    id: "ISS-{YYYYMMDD}-{counter:03d}",
-    title: "Auto-Test: " + failure.req_ref + " - " + failure.description,
-    status: "registered",
-    priority: severity_to_priority(failure.severity),
-    source: "auto-test",
-    phase_ref: PHASE_NUM,
-    description: "Expected: " + failure.expected + ". Actual: " + failure.actual,
-    fix_direction: failure.fix_suggestion.direction,
-    context: { location: failure.fix_suggestion.file + ":" + failure.fix_suggestion.line },
-    tags: ["auto-test", failure.layer]
-  }
-  Append to .workflow/issues/issues.jsonl
-```
-
-#### Report Display
-
-```
-=== AUTO-TEST RESULTS ===
-阶段:      {phase_name}
-来源:      {ROUTE}
-迭代:      {N} (策略: {strategy_history})
-收敛:      {status} ({final_pass_rate}%)
-
-层级结果:
-  L0 Static:      {pass_rate}% ({passed}/{total})
-  L1 Unit/API:    {pass_rate}% ({passed}/{total})
-  L2 Integration: {pass_rate}% ({passed}/{total})
-  L3 E2E:         {pass_rate}% ({passed}/{total})
-
-场景: {passed} passed, {failed} failed, {blocked} blocked
-Bugs: {N} discovered
-{IF spec: "需求覆盖: {pct}% | 已验证: {n}/{total}"}
-
-Files:
-  .tests/auto-test/state.json
-  .tests/auto-test/test-plan.json
-  .tests/auto-test/report.json
-  .tests/auto-test/reflection-log.md
-  {IF spec: ".tests/auto-test/traceability.md"}
-```
-
-#### Next-step routing
-
-| Result | Suggestion |
-|--------|------------|
-| Converged (>=threshold) | `/quality-review {phase}` to update validation |
-| All requirements verified (spec) | `/maestro-milestone-audit` |
-| Bugs discovered (code_defects) | `/quality-debug --from-auto-test {phase}` |
-| Max iter, >80% | `/quality-test {phase}` for manual UAT on remaining gaps |
-| Max iter, <80% | `/quality-debug {phase}` for deep investigation |
-| Coverage still low | `/quality-auto-test {phase} --layer {missing}` |
-| Re-run all pass | `/quality-review {phase}` |
-| Single pass (max_iter=1), bugs found | `/quality-debug --from-auto-test {phase}` |
-| Single pass, all pass | `/quality-test {phase}` |
+# Auto-Test Workflow (Unified)
+
+Unified automated testing with intelligent state-based routing. Merges test generation (gap-forward), business testing (PRD-forward), and integration testing (code-forward) into a single pipeline.
+
+**Core idea: read project state → auto-select scenario source → shared pipeline from plan through iteration.**
+
+Pipeline: Parse → Route → Source Scenarios → Discover Infrastructure → Plan → Write Tests → Execute → Iterate → Report
+Only Step 2 diverges (scenario source). Everything else is shared.
+
+---
+
+### Step 0: Parse Input & Load Specs
+
+**Parse arguments:**
+
+| Input | Result |
+|-------|--------|
+| No arguments | Error E001 |
+| Phase number | Resolve phase dir from artifact registry |
+| `--max-iter N` | Set MAX_ITER = N (default 5). **1 = single-pass** (no iteration loop) |
+| `--layer L` | Start from / restrict to specific layer (L0\|L1\|L2\|L3) |
+| `--dry-run` | Generate plan only, do not execute |
+| `--re-run` | Re-run only previously failed scenarios |
+
+**Resolve phase dir:** from `state.json` artifact registry (`type='execute'`, matching phase). Error E002 if not found.
+
+**Load specs:**
+```
+specs_test = maestro spec load --category test
+specs_arch = maestro spec load --category arch
+```
+
+`specs_test` for test conventions (Steps 3-5); `specs_arch` for module boundaries (Step 2).
+
+---
+
+### Step 1: Read State & Route
+
+Read project state signals and auto-select scenario source. This is the **primary branch point, with Route-specific extraction in Step 2** (after Step 2 normalization, the downstream pipeline is identical).
+
+```
+Priority: Resume > Re-run > Spec > Gap > Code
+
+1. RESUME:
+   Check: ${PHASE_DIR}/.tests/auto-test/state.json exists AND status == "running"
+   Compat: also check ${PHASE_DIR}/.tests/integration/state.json (old format)
+   Action: offer resume or restart
+   If resume: restore session, jump to Step 6 at saved iteration/layer
+   If restart: archive to .history/, continue as new
+
+2. RE-RUN:
+   Check: --re-run flag AND .tests/auto-test/report.json has failed/blocked scenarios
+   Compat: also check .tests/business/business-test-report.json (old format)
+   Action: load failed/blocked scenarios with status reset to pending
+   Skip to Step 4 (scenarios pre-loaded, plan confirmation)
+
+# Note: `integration/state.json` and `business/business-test-report.json` are
+# legacy compat paths read for backward compatibility only. The new pipeline
+# writes exclusively to `.tests/auto-test/state.json` and `.tests/auto-test/report.json`
+# (these legacy paths are superseded; see migration table at end of file).
+
+3. SPEC:
+   Check: .workflow/blueprint/BLP-*/requirements/REQ-*.md exists
+   Resolve: SPEC_DIR from index.json.blueprint_ref or most recent BLP-*/
+   If SPEC_DIR found: set ROUTE = "spec", SPEC_MODE = "full"
+   If no spec but has success_criteria: set ROUTE = "spec", SPEC_MODE = "degraded"
+
+4. GAP:
+   Check: ${PHASE_DIR}/verification.json has gaps[] with status MISSING or PARTIAL
+   OR: ${PHASE_DIR}/.tests/coverage-report.json has requirements_uncovered[]
+   Action: set ROUTE = "gap"
+
+5. CODE:
+   Default fallback when no spec package and no gaps detected.
+   Action: set ROUTE = "code"
+```
+
+**Display route selection:**
+```
+=== AUTO-TEST ===
+来源: {ROUTE} ({reason})
+阶段: {phase_name}
+{IF spec: "Spec: {SPEC_DIR} (mode: {SPEC_MODE})"}
+```
+
+---
+
+### Step 2: Source Scenarios
+
+Execute the route-specific scenario extraction, then normalize ALL scenarios into the unified format. **After this step, the pipeline is identical regardless of source.**
+
+#### Route A: `spec` source (PRD-forward)
+
+1. Load spec package:
+   - `requirements/REQ-*.md` — functional requirements with acceptance criteria
+   - `requirements/NFR-*.md` — non-functional requirements
+   - `architecture/_index.md` — API endpoints, data model, state machines
+   - `epics/EPIC-*.md` — user stories for E2E scenarios
+
+2. For each `REQ-NNN-{slug}.md`, parse `## Acceptance Criteria`:
+   - Extract each numbered criterion as a scenario seed
+   - Map RFC 2119 keywords to priority:
+
+   | Keyword | Priority | Failure Severity |
+   |---------|----------|-----------------|
+   | MUST / MUST NOT / SHALL / SHALL NOT | critical | blocker |
+   | SHOULD / SHOULD NOT / RECOMMENDED | high | major |
+   | MAY / OPTIONAL | medium | minor |
+
+3. Classify scenario into layer:
+
+   | Source | Layer | Category |
+   |--------|-------|----------|
+   | Architecture API endpoints + REQ AC about request/response | L1 | api_contract |
+   | REQ AC about business logic, validation, state changes | L2 | business_rule |
+   | REQ AC about state transitions (from architecture state machines) | L2 | state_transition |
+   | Epic user stories (multi-step flows) | L3 | user_flow |
+   | NFR performance/security constraints | L2 | non_functional |
+
+4. **Generate Fixtures** (conditional sub-step):
+
+   **Tier 1: Schema-derived fixtures**
+   ```
+   FOR each entity in REQ data model:
+     valid: object satisfying all field constraints
+     invalid: one variant per constraint violation (null_required, empty_string, overflow, wrong_type)
+     boundary: edge value variants (min_value, max_value, min_minus_one, max_plus_one)
+   ```
+
+   **Tier 2: Acceptance-criteria-derived expectations**
+   ```
+   FOR each "MUST return X when Y": fixture { input: Y, expected: X }
+   FOR each "MUST validate Z": fixture { input: invalid_Z, expected: validation_error }
+   ```
+
+   **Tier 3: Business-scenario-derived data sets (L3 only)**
+   ```
+   FOR each Epic user story:
+     scenario_pack: coordinated data set across story steps
+   ```
+
+   **Mock contracts:**
+   ```
+   FOR each API endpoint in architecture/_index.md:
+     mock_contract: { request_pattern, response_fixture }
+   ```
+
+5. **Degraded mode** (SPEC_MODE = "degraded"):
+   - Extract from `index.json.success_criteria` (each → one L2 scenario)
+   - Extract from `plan.json` task convergence criteria (each → one L1/L2 scenario)
+   - Extract from `.summaries/TASK-*-summary.md` (each → one L1 scenario)
+   - All default to priority: "high". No L3 in degraded mode.
+
+6. Convert to unified scenario format.
+
+#### Route B: `gap` source (coverage-forward)
+
+1. Read gap sources:
+   - `verification.json` → `gaps[]` with status MISSING or PARTIAL
+   - `coverage-report.json` → `requirements_uncovered[]`
+   - Task summaries → modified files list
+
+2. Priority: MISSING or uncovered → HIGH; PARTIAL → MEDIUM.
+
+3. Classify each changed file:
+
+   | File Type | Category | Layer |
+   |-----------|----------|-------|
+   | Pure function / utility | unit | L1 |
+   | React component | unit + e2e | L1 + L3 |
+   | API route / handler | integration | L2 |
+   | Database model / query | integration | L2 |
+   | CLI command | e2e | L3 |
+   | Config / types / constants / CSS / test files | skip | — |
+
+4. **Optional: CLI supplementary analysis** (skip if no CLI tools enabled or all files "skip"):
+   ```
+   Bash({
+     command: 'maestro delegate "PURPOSE: Analyze source files for test-worthy edge cases
+   TASK: Identify error handling | boundary conditions | state transitions | external dependencies
+   MODE: analysis
+   CONTEXT: @${target_files}
+   EXPECTED: JSON array of { file, edge_cases: [{ description, type, priority }] }
+   CONSTRAINTS: Non-obvious cases only | Max 5 per file
+   " --role analyze --mode analysis',
+     run_in_background: true
+   })
+   ```
+   On callback: merge edge_cases into scenarios, mark `source: "cli-analysis"`.
+
+5. Convert to unified scenario format.
+
+#### Route C: `code` source (exploration-forward)
+
+1. Explore codebase for testable integration points:
+   - Module boundaries and cross-module calls
+   - API endpoints and their handlers
+   - Database interactions and queries
+   - External service integrations
+   - Event flows and message passing
+
+2. Scan for cross-module imports, API route definitions, database calls.
+
+3. Map integration points: which modules communicate through what interfaces.
+
+4. Infer layer from integration type:
+   - Isolated function → L1
+   - Cross-module call / API handler → L2
+   - Full user flow → L3
+
+5. Convert to unified scenario format.
+
+#### Unified Scenario Format
+
+All routes produce this identical structure:
+
+```json
+{
+  "id": "AT-{NNN}",
+  "source": "spec|gap|code|re-run",
+  "layer": "L0|L1|L2|L3",
+  "priority": "critical|high|medium",
+  "category": "api_contract|business_rule|state_transition|user_flow|non_functional|unit|integration|e2e|static",
+  "name": "descriptive scenario name",
+  "target_file": "src/path/to/file.ts",
+  "test_file": "src/path/__tests__/file.test.ts",
+  "req_ref": "REQ-NNN:AC-N | gap-id | null",
+  "description": "what this scenario validates",
+  "test_cases": ["case 1", "case 2"],
+  "input": { "$fixture_ref": "..." },
+  "expected": {
+    "status": 200,
+    "behavior": "description of expected outcome"
+  },
+  "preconditions": [],
+  "postconditions": [],
+  "mock_services": [],
+  "fixtures": {}
+}
+```
+
+**Field population by route:**
+
+| Field | spec | gap | code |
+|-------|------|-----|------|
+| `req_ref` | REQ-NNN:AC-N | gap ID from verification.json | null |
+| `fixtures` | Tier 1-3 generated | empty (inferred in Step 5) | empty (inferred in Step 5) |
+| `mock_services` | from architecture/_index.md | empty | discovered from imports |
+| `preconditions` | from REQ AC text | empty | empty |
+
+---
+
+### Step 3: Discover Test Infrastructure
+
+Detect existing test framework and patterns:
+
+1. **Config files**: `jest.config.*`, `vitest.config.*`, `pytest.ini`, `pyproject.toml`, `.mocharc.*`
+2. **Existing tests**: `*.test.*`, `*.spec.*`, `test_*` (exclude node_modules, .git)
+3. **Utilities**: `test-utils.*`, `testHelper*`, `conftest.py`, `setup.*`
+
+Extract: framework, directory structure, naming convention, test utilities, run command.
+
+Read 2-3 existing test files to learn: import style, describe/it nesting, assertion library, mock patterns, setup/teardown.
+
+**Detect tech stack for generation:**
+
+| Stack | L1 | L2 | L3 |
+|-------|----|----|-----|
+| Java/Spring Boot | MockMvc | JUnit 5 + WireMock | TestContainers |
+| TypeScript/Node | vitest | supertest + nock | playwright/cypress |
+| Python | pytest | httpx + responses | selenium |
+
+If no test framework detected: Error E003.
+
+Output: `infrastructure` object passed to Steps 5-6.
+
+---
+
+### Step 4: Generate Test Plan & Confirm
+
+1. **Merge pre-existing tests** from `.tests/test-gen-report.json` (if exists):
+   - Mark as "pre-existing" so Step 5 skips writing them
+   - Step 6 includes them in execution
+
+2. Apply `--layer` filter if specified.
+
+3. **Archive previous plan** to `.history/` if exists.
+
+4. Write `.tests/auto-test/test-plan.json`:
+```json
+{
+  "phase": "{phase}",
+  "source_route": "{ROUTE}",
+  "blueprint_ref": "{SPEC_DIR name or null}",
+  "spec_mode": "full|degraded|null",
+  "generated_at": "{ISO timestamp}",
+  "infrastructure": { "framework": "...", "run_command": "..." },
+  "layers": {
+    "L0": { "scenario_count": N, "commands": ["tsc --noEmit", "eslint src/"] },
+    "L1": { "scenario_count": N, "priority_distribution": { "critical": X, "high": Y, "medium": Z } },
+    "L2": { "scenario_count": N, "priority_distribution": { ... } },
+    "L3": { "scenario_count": N, "priority_distribution": { ... } }
+  },
+  "scenarios": [ "...unified format..." ],
+  "fixtures": { "...if spec source..." },
+  "requirement_coverage_plan": { "requirements_targeted": [], "requirements_skipped": [] }
+}
+```
+
+5. Display and confirm:
+```
+=== AUTO-TEST PLAN ===
+来源:  {ROUTE}
+阶段:  {phase_name}
+Blueprint: {blueprint_ref or "N/A"}
+
+层级分布:
+  L0 Static:      {N} checks
+  L1 Unit/API:    {N} scenarios ({X} critical, {Y} high)
+  L2 Integration: {N} scenarios ({X} critical, {Y} high)
+  L3 E2E:         {N} scenarios ({X} critical, {Y} high)
+
+Total: {N} scenarios, {M} test cases
+Max iterations: {max_iter}
+
+Proceed? (yes/edit/cancel)
+```
+
+- `--dry-run`: stop here, report plan
+- User "edit": modify plan interactively
+- User "cancel": abort
+
+---
+
+### Step 5: Write Tests (RED-GREEN) via CSV Parallel
+
+**Parallel strategy**: Build `layer-L{N}-write.csv` per layer, execute via `spawn_agents_on_csv`. Each agent writes one test file independently.
+
+#### 5a. Build Write CSV
+
+For each layer (L1, L2, L3 — sequential):
+
+```
+Extract pending scenarios for this layer from test-plan.json
+Build layer-L{N}-write.csv:
+  Columns: id, name, target_file, test_file, description, test_cases, fixtures, req_ref, infrastructure_hints, prev_context
+
+  prev_context = findings from completed prior-layer scenarios (cross-layer propagation)
+```
+
+#### 5b. Parallel Test Writing via spawn_agents_on_csv
+
+```javascript
+spawn_agents_on_csv({
+  csv_path: `.tests/auto-test/.csv-session/layer-L${N}-write.csv`,
+  id_column: "id",
+  instruction: `
+    Write ONE test file for the given scenario using RED-GREEN methodology.
+
+    Rules:
+    - Read target_file to understand module under test
+    - Write test at test_file path following infrastructure_hints patterns
+    - Each test_case → one it() block, include id in describe("AT-NNN: {name}")
+    - Use fixtures column (infer from source if empty)
+    - Run test once after writing: report red_result
+    - NEVER modify source code — only write/fix test files
+
+    RED results:
+    - pass: test passes immediately (may need strengthening)
+    - expected_fail: test correctly targets real behavior
+    - unexpected_fail: setup/import error — fix test, re-run
+
+    Read discoveries.ndjson for shared patterns. Append if you find reusable ones.
+  `,
+  max_concurrency: 5,
+  max_runtime_seconds: 1800,
+  output_csv_path: `.tests/auto-test/.csv-session/layer-L${N}-write-results.csv`,
+  output_schema: { id, status: [written|failed], red_result: [expected_fail|unexpected_fail|pass], findings, error }
+})
+```
+
+#### 5c. Merge & Continue
+
+Merge write-results into master state. Delete temp CSV. Proceed to next layer or Step 6.
+
+**If `--max-iter 1`:** After all layers written and run once, jump directly to Step 8 (single-pass mode, replaces test-gen behavior). Skip Steps 6-7.
+
+---
+
+### Step 6: Execute (Progressive Layers)
+
+Run tests progressively through layers with fail-fast on critical:
+
+- L0: `tsc --noEmit` + `eslint src/` (static analysis)
+- L1: unit tests (`--testPathPattern="unit|__tests__"`)
+- L2: integration tests (`--testPathPattern="integration"`)
+- L3: E2E tests (`--testPathPattern="e2e"`)
+
+**Fail-fast rule:**
+- L0 must pass before L1
+- If ANY "critical" priority failure in L1 → do NOT proceed to L2
+- If ANY "critical" priority failure in L2 → do NOT proceed to L3
+- "high" and "medium" failures do NOT block next layer
+
+Record per-scenario results: `{ status, actual_response, duration_ms, error_detail, classification: null }`.
+
+Write iteration results to `.tests/auto-test/results-iter-{N}.json`.
+
+---
+
+### Step 7: Reflect & Adjust (Unified Iteration Engine)
+
+Single engine that subsumes both Generator-Critic (per-layer inner) and Reflect-Adjust (global outer). Uses `spawn_agents_on_csv` for parallel failure diagnosis.
+
+```
+OUTER LOOP (max_iter iterations):
+
+  FOR each active layer (L0 through current):
+
+    INNER LOOP (max 3 iterations per layer — Generator-Critic):
+
+      1. Build diagnosis CSV from failed scenarios:
+
+         Build diagnosis-iter-{N}.csv:
+           Columns: id, scenario_id, layer, test_file, error_detail, expected, actual, target_file, source_context
+
+         spawn_agents_on_csv({
+           csv_path: `.tests/auto-test/.csv-session/diagnosis-iter-${iter}.csv`,
+           id_column: "id",
+           instruction: `
+             Classify ONE test failure and provide fix if applicable.
+
+             Classifications:
+             - test_defect: Test wrong (bad import, endpoint, fixture, assertion)
+             - code_defect: Source violates business rule (actual != expected)
+             - env_issue: Environment problem (service down, config missing)
+
+             If test_defect: provide fix_code (old → new)
+             If code_defect/env_issue: leave fix_code empty, provide evidence
+
+             Rules:
+             - NEVER suggest source code changes
+             - A test correctly catching a real bug = code_defect, not test_defect
+             - When uncertain: prefer code_defect (conservative)
+           `,
+           max_concurrency: 5,
+           max_runtime_seconds: 1200,
+           output_csv_path: `.tests/auto-test/.csv-session/diagnosis-iter-${iter}-results.csv`,
+           output_schema: { id, classification, fix_code, evidence, error }
+         })
+
+      2. Merge diagnosis results:
+         - test_defect with fix_code → apply fix to test file
+         - code_defect → mark as confirmed failure (stop retrying)
+         - env_issue → mark as blocked
+
+      3. IF test_defects found:
+         Re-run ALL scenarios in this layer (catch regressions)
+
+      4. IF no test_defects remain: break inner loop
+
+    END INNER
+
+    Record final layer results
+    IF critical code_defects in this layer: stop layer progression (fail-fast)
+
+  END FOR
+
+  REFLECT:
+    Analyze: which tests failed, pass rate improving/plateauing/regressing,
+    failures clustered by component, strategy effectiveness.
+    Append to reflection-log.md:
+      iteration, strategy, pass_rate delta, what worked/failed,
+      detected patterns, strategy assessment (effective/ineffective + recommendation)
+
+  ADJUST (Adaptive Strategy Engine):
+
+    | Condition | Strategy | Behavior |
+    |-----------|----------|----------|
+    | Iteration 1-2 | Conservative | Fix obvious test_defects only, don't refactor |
+    | Pass rate >80% AND failures similar | Aggressive | Batch-fix related failures together |
+    | New regressions appeared | Surgical | Revert last changes, fix regression only |
+    | Stuck 3+ iterations (rate not improving) | Reflective | Step back, re-analyze root cause pattern |
+
+    Transitions:
+      Conservative --(>80%)--> Aggressive
+      Aggressive --(regression)--> Surgical --(fixed)--> Aggressive
+      Any --(stuck 3+ iters)--> Reflective --(insight)--> Conservative
+
+  CONVERGENCE CHECK:
+    pass_rate >= threshold (95%) → Step 8 (converged)
+    iteration >= max_iter → Step 8 (max_iter_reached)
+    all remaining failures = code_defect → Step 8 (confirmed_defects)
+    ELSE → next outer iteration (back to Execute)
+
+END OUTER
+```
+
+**Degenerate cases:**
+- `max_iter=1`: Step 5 writes tests, Step 6 executes once, Step 7 runs Reflect only (log results), no Adjust, no loop → Step 8. Equivalent to test-gen single pass.
+- `max_iter=3`: Inner loop cleans test_defects. Up to 3 outer passes. Similar to business-test.
+- `max_iter=5`: Full adaptive strategy progression. Similar to integration-test.
+
+---
+
+### Step 7.5: Test Confidence Scoring
+
+Scored after each REFLECT step. Dimensions (5): scenario_coverage, test_quality, diagnostic_accuracy, strategy_effectiveness, infrastructure_fitness. Factors (weights): completeness(.30), pass_rate_trend(.25), classification_accuracy(.20), coverage_breadth(.15), consistency(.10). Append confidence table to reflection-log.md.
+
+**Enhanced Convergence**: pass_rate ≥ 95% AND confidence ≥ 60% → converged. pass_rate ≥ 95% BUT confidence < 60% → continue (tests may be weak). max_iter reached or all failures = code_defect → Step 8.
+
+**Quality mechanisms**: Pressure Pass (before Step 8) — select 2-3 passing tests from highest-pass-rate layer, verify they exercise real behavior (not mock-only, non-trivial assertions). Devil's Advocate — pass_rate > 80% → challenge assertion specificity, error path coverage, mock over-reliance. Stall Detection — delta < 5% for 2 iterations + pass_rate flat → force Reflective strategy.
+
+**Readiness Gate** (before Step 8, skip if max_iter=1): scenario_coverage < 40% | no pressure pass | diagnostic_accuracy < 40% | unclassified failures. If blocked → force one additional iteration. Add confidence section to report.json.
+
+---
+
+### Step 8: Complete & Write Artifacts
+
+1. Update session state:
+```json
+// .tests/auto-test/state.json
+{
+  "session_id": "auto-test-{YYYYMMDD-HHmmss}",
+  "phase": "{phase}",
+  "phase_dir": "{PHASE_DIR}",
+  "source_route": "spec|gap|code|re-run",
+  "blueprint_ref": "SPEC-001 | null",
+  "spec_mode": "full|degraded|null",
+  "status": "converged|max_iter_reached|confirmed_defects|single_pass",
+  "flags": { "max_iter": 5, "layer": null, "dry_run": false, "re_run": false },
+  "iteration": 2,
+  "strategy": "conservative",
+  "strategy_history": ["conservative", "aggressive"],
+  "threshold": 95,
+  "current_layer": "L2",
+  "layer_state": {
+    "L0": { "inner_iter": 1, "pass_rate": 100.0, "status": "passed" },
+    "L1": { "inner_iter": 2, "pass_rate": 95.0, "status": "passed" },
+    "L2": { "inner_iter": 1, "pass_rate": 87.5, "status": "completed" },
+    "L3": { "inner_iter": 0, "pass_rate": 0, "status": "pending" }
+  },
+  "pass_rate_history": [72.0, 85.0, 95.0],
+  "scenario_count": 30,
+  "infrastructure": { "framework": "vitest", "run_command": "npm test" },
+  "started_at": "{ISO}",
+  "updated_at": "{ISO}"
+}
+```
+
+2. Archive previous report to `.history/` if exists.
+
+3. Write `.tests/auto-test/report.json`:
+```json
+{
+  "phase": "{phase}",
+  "source_route": "{ROUTE}",
+  "blueprint_ref": "{spec ref or null}",
+  "spec_mode": "full|degraded|null",
+  "completed_at": "{ISO timestamp}",
+  "convergence": {
+    "status": "converged|max_iter_reached|confirmed_defects|single_pass",
+    "iterations": N,
+    "final_pass_rate": 95.0,
+    "threshold": 95,
+    "strategy_history": ["conservative", "aggressive"]
+  },
+  "infrastructure": { "framework": "vitest", "test_dir": "__tests__/", "run_command": "npm test" },
+  "layers": {
+    "L0": { "total": N, "passed": P, "failed": F, "blocked": B, "pass_rate": 100.0 },
+    "L1": { "total": N, "passed": P, "failed": F, "blocked": B, "pass_rate": 95.0 },
+    "L2": { ... },
+    "L3": { ... }
+  },
+  "scenarios": [
+    {
+      "id": "AT-001", "source": "spec", "layer": "L1", "name": "...",
+      "test_file": "...", "req_ref": "REQ-001:AC-1",
+      "status": "passed|failed|blocked",
+      "classification": "null|test_defect|code_defect|env_issue",
+      "iterations_to_pass": 2
+    }
+  ],
+  "failures": [
+    {
+      "id": "AF-001", "scenario_id": "AT-005", "req_ref": "REQ-001:AC-3",
+      "layer": "L1", "severity": "critical", "classification": "code_defect",
+      "description": "...", "expected": "...", "actual": "...",
+      "fix_suggestion": { "file": "src/...", "line": 42, "direction": "..." }
+    }
+  ],
+  "requirement_coverage": [],
+  "summary": {
+    "source_route": "{ROUTE}",
+    "total_scenarios": 30, "total_passed": 26, "total_failed": 3, "total_blocked": 1,
+    "bugs_discovered": 3, "test_defects_fixed": 5, "coverage_pct": 85.0
+  }
+}
+```
+
+4. Update `validation.json`: if gap source, change MISSING → COVERED for passing tests.
+5. Update `index.json` with `auto_test` section.
+
+**Artifact file mapping (old → new):**
+
+| Old file | New file |
+|----------|----------|
+| `.tests/test-gen-report.json` | `.tests/auto-test/report.json` (source_route: "gap") |
+| `.tests/integration/state.json` | `.tests/auto-test/state.json` |
+| `.tests/integration/summary.json` | Merged into report.json |
+| `.tests/integration/reflection-log.md` | `.tests/auto-test/reflection-log.md` |
+| `.tests/business/business-test-report.json` | `.tests/auto-test/report.json` (source_route: "spec") |
+| `.tests/business/business-test-summary.md` | `.tests/auto-test/traceability.md` (conditional) |
+
+---
+
+### Step 9: Post-Processing & Routing
+
+#### Conditional: Traceability Matrix (spec source only)
+
+Build REQ → AC → scenario → result mapping:
+
+```
+FOR each REQ in requirements targeted:
+  FOR each AC in REQ:
+    scenarios_for_ac = filter scenarios where req_ref == "REQ-NNN:AC-{N}"
+    ac_status = "passed" if ALL passed
+                "failed" if ANY failed
+                "blocked" if ANY blocked and none failed
+                "untested" if no scenarios mapped
+
+  req.coverage_pct = passed_criteria / total_criteria * 100
+  req.verdict = "verified" if all MUST+SHOULD passed
+                "partial" if some failed
+                "unverified" if all failed or untested
+```
+
+Populate `requirement_coverage[]` in report.json.
+Write `.tests/auto-test/traceability.md` (human-readable table).
+
+#### Conditional: Issue Creation (when failures exist)
+
+```
+FOR each failure in report.failures WHERE classification == "code_defect":
+  issue = {
+    id: "ISS-{YYYYMMDD}-{counter:03d}",
+    title: "Auto-Test: " + failure.req_ref + " - " + failure.description,
+    status: "registered",
+    priority: severity_to_priority(failure.severity),
+    source: "auto-test",
+    phase_ref: PHASE_NUM,
+    description: "Expected: " + failure.expected + ". Actual: " + failure.actual,
+    fix_direction: failure.fix_suggestion.direction,
+    context: { location: failure.fix_suggestion.file + ":" + failure.fix_suggestion.line },
+    tags: ["auto-test", failure.layer]
+  }
+  Append to .workflow/issues/issues.jsonl
+```
+
+#### Report Display
+
+```
+=== AUTO-TEST RESULTS ===
+阶段:      {phase_name}
+来源:      {ROUTE}
+迭代:      {N} (策略: {strategy_history})
+收敛:      {status} ({final_pass_rate}%)
+
+层级结果:
+  L0 Static:      {pass_rate}% ({passed}/{total})
+  L1 Unit/API:    {pass_rate}% ({passed}/{total})
+  L2 Integration: {pass_rate}% ({passed}/{total})
+  L3 E2E:         {pass_rate}% ({passed}/{total})
+
+场景: {passed} passed, {failed} failed, {blocked} blocked
+Bugs: {N} discovered
+{IF spec: "需求覆盖: {pct}% | 已验证: {n}/{total}"}
+
+Files:
+  .tests/auto-test/state.json
+  .tests/auto-test/test-plan.json
+  .tests/auto-test/report.json
+  .tests/auto-test/reflection-log.md
+  {IF spec: ".tests/auto-test/traceability.md"}
+```
+
+#### Next-step routing
+
+| Result | Suggestion |
+|--------|------------|
+| Converged (>=threshold) | `/quality-review {phase}` to update validation |
+| All requirements verified (spec) | `/maestro-milestone-audit` |
+| Bugs discovered (code_defects) | `/quality-debug --from-auto-test {phase}` |
+| Max iter, >80% | `/quality-test {phase}` for manual UAT on remaining gaps |
+| Max iter, <80% | `/quality-debug {phase}` for deep investigation |
+| Coverage still low | `/quality-auto-test {phase} --layer {missing}` |
+| Re-run all pass | `/quality-review {phase}` |
+| Single pass (max_iter=1), bugs found | `/quality-debug --from-auto-test {phase}` |
+| Single pass, all pass | `/quality-test {phase}` |