QA Party

QA_PARTY Skill

Purpose: Coordinated parallel validation with 8+ QA agents Created: 2025-12-31 Trigger: /qa-party command Aliases: /validate, /qa, /qp

When to Use

Deploy QA_PARTY when you need comprehensive validation:

• After CCW completes a workstream
• Before creating a PR
• After merging changes
• Regression testing after major refactors
• CI/CD equivalent for local validation

Do NOT use for:

• Single test file execution
• Quick lint check
• When you know exactly what failed

Economics: Zero Marginal Wall-Clock Cost

Critical Understanding: Parallel agents with the same timeout cost nothing extra in wall-clock time.

Sequential (BAD):        Parallel (GOOD):
8 checks × 60s each      8 checks × 60s in parallel
Total: 480s              Total: 60s (8x faster)

Implication: Always spawn all QA agents. There is no cost savings from running fewer.

Deployment Pattern

Standard Deployment: 8 QA Agents

Deploy 8 QA agents in parallel, each running a specific validation domain:

Extended Deployment: 12 QA Agents

For comprehensive validation, add domain-specific test splits:

Invocation

Full Deployment (8 agents)

/qa-party

Deploys all 8 QA agents in parallel.

Quick Deployment (4 agents)

/qa-party --quick

Deploys 4 critical agents (BUILD, LINT-BACKEND, LINT-FRONTEND, DOCKER).

Extended Deployment (12 agents)

/qa-party --full

Deploys all 12 agents including domain-specific test splits.

Single Domain

/qa-party frontend
/qa-party backend
/qa-party docker

Deploys agents for specific domain only.

Spawn Pattern

# Deploy all 8 QA agents in parallel
# Wall-clock = slowest agent timeout

spawn_parallel([
    Task(subagent_type="general-purpose", description="QA-FRONTEND-BUILD",
         prompt="""
Run frontend build validation:
```bash
cd /Users/aaronmontgomery/Autonomous-Assignment-Program-Manager/frontend
npm run build 2>&1

Report: PASS/FAIL with error count and key errors if any. """),

Task(subagent_type="general-purpose", description="QA-FRONTEND-LINT",
     prompt="""

Run frontend lint validation:

cd /Users/aaronmontgomery/Autonomous-Assignment-Program-Manager/frontend
npm run lint 2>&1
npm run type-check 2>&1

Report: PASS/FAIL with error count and categories. """),

Task(subagent_type="general-purpose", description="QA-FRONTEND-TESTS",
     prompt="""

Run frontend test suite:

cd /Users/aaronmontgomery/Autonomous-Assignment-Program-Manager/frontend
npm test -- --passWithNoTests 2>&1

Report: PASS/FAIL with test counts (passed/failed/skipped). """),

Task(subagent_type="general-purpose", description="QA-BACKEND-LINT",
     prompt="""

Run backend lint validation:

cd /Users/aaronmontgomery/Autonomous-Assignment-Program-Manager/backend
./venv/bin/ruff check . 2>&1
./venv/bin/ruff format --check . 2>&1

Report: PASS/FAIL with error count and categories. """),

Task(subagent_type="general-purpose", description="QA-BACKEND-TESTS",
     prompt="""

Run backend test suite:

cd /Users/aaronmontgomery/Autonomous-Assignment-Program-Manager/backend
./venv/bin/python -m pytest tests/ -v --tb=short 2>&1 | head -100

Report: PASS/FAIL with test counts (passed/failed/errors). """),

Task(subagent_type="general-purpose", description="QA-DOCKER-HEALTH",
     prompt="""

Check Docker container health:

docker compose ps 2>&1
docker compose logs --tail=20 backend 2>&1
docker compose logs --tail=20 mcp-server 2>&1

Report: Container status (running/stopped) and any error logs. """),

Task(subagent_type="general-purpose", description="QA-API-SMOKE",
     prompt="""

Run API smoke tests:

curl -s http://localhost:8000/health 2>&1
curl -s http://localhost:8000/api/v1/docs 2>&1 | head -20

Report: PASS/FAIL with response status and any errors. """),

Task(subagent_type="general-purpose", description="QA-MCP-TOOLS",
     prompt="""

Validate MCP server tools:

docker compose exec mcp-server python -c "from scheduler_mcp.server import mcp; print(f'Tools loaded: {len(mcp.tools)}')" 2>&1

Report: PASS/FAIL with tool count. """), ])

---

## Output Format

### Per-Agent Report

Each QA agent reports:

```markdown
## QA Agent: [NAME]

**Status:** PASS / FAIL / PARTIAL
**Duration:** Xs

### Results
- [Metric 1]: [value]
- [Metric 2]: [value]

### Errors (if any)

[error output]

### Recommended Fix
[If FAIL, suggest fix]

Consolidated Dashboard

## QA_PARTY Validation Dashboard

| Agent | Status | Duration | Key Metric |
|-------|--------|----------|------------|
| QA-FRONTEND-BUILD | PASS | 45s | 0 errors |
| QA-FRONTEND-LINT | FAIL | 12s | 6 errors |
| QA-FRONTEND-TESTS | FAIL | 8s | 164 failing |
| QA-BACKEND-LINT | PASS | 5s | 0 errors |
| QA-BACKEND-TESTS | PASS | 120s | 312/312 passed |
| QA-DOCKER-HEALTH | PASS | 3s | 7/7 running |
| QA-API-SMOKE | PASS | 1s | 200 OK |
| QA-MCP-TOOLS | PASS | 2s | 70 tools |

### Overall: 6/8 PASS (75%)

### Blocking Issues
1. Frontend lint: 6 errors (`no-this-alias` × 5, `no-assign-module-variable` × 1)
2. Frontend tests: 164 failing (Jest transform broken)

### Non-Blocking Issues
[None]

### Recommended CCW Tasks
1. Fix `no-this-alias` violations in `src/lib/validation/error-messages.ts`
2. Fix Jest transform configuration

Failure Recovery

Minimum Viable QA

Validation can proceed with degraded confidence if:

• QA-BACKEND-LINT passes ✓
• QA-BACKEND-TESTS passes ✓
• QA-DOCKER-HEALTH passes ✓

Circuit Breaker

If Docker is down: Skip QA-API-SMOKE, QA-MCP-TOOLS, QA-BACKEND-TESTS (containerized).

Integration with CCW

Pre-CCW Validation

/qa-party --quick

Establish baseline before CCW burns tasks.

Post-CCW Validation

/qa-party --full

Comprehensive validation after CCW completes.

Report to CCW

After running /qa-party, generate CCW directive:

## CCW Task from QA_PARTY

QA-FRONTEND-LINT failed with 6 errors:

Files to fix:
- `frontend/src/lib/validation/error-messages.ts` (lines 18, 40, 68, 97, 315)
- `frontend/src/utils/debounce.ts` (no-this-alias)
- `frontend/src/utils/lazy-loader.ts` (no-assign-module-variable)

Pattern to apply:
```typescript
// Before (triggers no-this-alias)
const self = this;
callback.apply(self, args);

// After
callback.apply(this, args);
// OR use arrow function

Acceptance: npm run lint exits 0.

---

## Related Skills

| Skill | When to Use |
|-------|-------------|
| `search-party` | Reconnaissance before QA (G-2 function) |
| `code-review` | IG teams invoke for code quality checks |
| `systematic-debugger` | When QA fails, debug root cause |
| `test-writer` | When QA reveals missing coverage |
| `lint-monorepo` | Unified linting for Python/TypeScript |
| `security-audit` | Security-focused code audit |

### Skill vs Subagent Types

**Important distinction:**

| Type | Mechanism | Examples |
|------|-----------|----------|
| **Skills** | Invoke via Skill tool | `/code-review`, `/test-writer`, `/lint-monorepo` |
| **Subagents** | Spawn via Task tool | `Explore`, `Plan`, `general-purpose` |

IG teams use `subagent_type="general-purpose"` with Task tool, but they CAN invoke skills like `/code-review` within their execution.

---

## Command Structure

### Oversight Hierarchy

QA/Testing falls under **IG (Inspector General)** function, not G-2 (Intel):

ORCHESTRATOR (you) └── COORD_QUALITY (IG Commander - Inspections & Quality) ├── QA-IG-BACKEND-ROUTES (10 test agents) ├── QA-IG-BACKEND-SERVICES (10 test agents) ├── QA-IG-BACKEND-MODELS (10 test agents) ├── QA-IG-BACKEND-SCHEDULING (10 test agents) ├── QA-IG-BACKEND-RESILIENCE (10 test agents) ├── QA-IG-BACKEND-ACGME (10 test agents) ├── QA-IG-FRONTEND-COMPONENTS (10 test agents) ├── QA-IG-FRONTEND-HOOKS (10 test agents) ├── QA-IG-FRONTEND-PAGES (10 test agents) ├── QA-IG-LINT-BACKEND (10 lint agents) ├── QA-IG-LINT-FRONTEND (10 lint agents) └── QA-IG-SECURITY (10 audit agents) = 120 QA agents

**Staff Distinction:**
- **G-2 (SEARCH_PARTY)**: Intelligence/Reconnaissance - explores and gathers information
- **IG (QA_PARTY)**: Inspector General - tests, validates, and ensures quality

### Coordinator Responsibilities

| Role | Agent | Responsibility |
|------|-------|----------------|
| **IG Commander** | COORD_QUALITY | Deploys 12 IG teams, synthesizes results, assigns follow-up |
| **IG Teams** | QA-IG-* (12) | Each manages 10 test workers, reports to COORD_QUALITY |
| **Workers** | QA agents (120) | Execute single test file/slice, report PASS/FAIL |
| **Auditor** | DELEGATION_AUDITOR | Post-run metrics (agent efficiency, failure patterns) |
| **AAR** | COORD_AAR | After Action Review (what went well, what didn't) |

### Coordinator Spawn Pattern

```python
# ORCHESTRATOR spawns COORD_QUALITY who spawns 12 G-2s
Task(
    subagent_type="general-purpose",
    description="COORD_QUALITY: QA_PARTY Commander",
    prompt="""
## Agent: COORD_QUALITY
You are the Quality Coordinator commanding QA_PARTY deployment.

## Mission
Deploy 12 IG QA teams in parallel. Each team runs 10 test workers.
Collect all reports and synthesize into unified dashboard.

## Your IG Teams to Deploy
1. QA-IG-BACKEND-ROUTES
2. QA-IG-BACKEND-SERVICES
3. QA-IG-BACKEND-MODELS
4. QA-IG-BACKEND-SCHEDULING
5. QA-IG-BACKEND-RESILIENCE
6. QA-IG-BACKEND-ACGME
7. QA-IG-FRONTEND-COMPONENTS
8. QA-IG-FRONTEND-HOOKS
9. QA-IG-FRONTEND-PAGES
10. QA-IG-LINT-BACKEND
11. QA-IG-LINT-FRONTEND
12. QA-IG-SECURITY

## Spawn all 12 in parallel using Task tool with subagent_type="general-purpose"

## After all report back:
1. Aggregate results into dashboard
2. Identify blocking issues
3. Generate CCW task list for failures
4. Report to ORCHESTRATOR
"""
)

IG Team Spawn Pattern (Each Team Deploys 10 Workers)

# Example: QA-IG-BACKEND-ROUTES spawning 10 workers
Task(
    subagent_type="general-purpose",
    description="QA-IG-BACKEND-ROUTES",
    prompt="""
## Agent: QA-IG-BACKEND-ROUTES
You are an IG QA team commanding 10 test workers for backend routes.

## Mission
Run pytest on route test files. Distribute across 10 parallel workers.

## Your Test Slice
```bash
cd /Users/aaronmontgomery/Autonomous-Assignment-Program-Manager/backend
find tests -name "test_*routes*.py" -o -name "test_*api*.py"

Spawn 10 workers (or use pytest-xdist)

Option A: Task tool with 10 parallel agents Option B: pytest -n 10 --dist loadfile tests/routes/

Report Format

Report to COORD_QUALITY with this table. """ )

---

## Scaled Deployment: 120 QA Agents

### How to Get 120 QA Testers Going

The key is **test file distribution**. Instead of running one agent per validation type, we split the work:

| Validation Type | Approximate Files | Agents |
|-----------------|-------------------|--------|
| Backend test files | ~60 files | 60 agents |
| Frontend test files | ~50 files | 50 agents |
| Frontend components (untested) | ~254 | 10 agents (batch) |
| **Total** | | **120 agents** |

### Invocation

/qa-party --scale 120

### Discovery Phase (Run First)

Before spawning 120 agents, discover test targets:

```bash
# Backend: Find all test files
find backend/tests -name "test_*.py" -type f | wc -l  # ~60 files

# Frontend: Find all test files
find frontend/__tests__ -name "*.test.ts" -o -name "*.test.tsx" | wc -l  # ~50 files

# Frontend: Find untested components
find frontend/src/components -name "*.tsx" | wc -l  # ~254 components

Scaled Spawn Pattern

# Phase 1: Discover test files
import subprocess

# Backend tests
backend_tests = subprocess.run(
    ["find", "backend/tests", "-name", "test_*.py", "-type", "f"],
    capture_output=True, text=True
).stdout.strip().split("\n")

# Frontend tests
frontend_tests = subprocess.run(
    ["find", "frontend/__tests__", "-name", "*.test.ts*"],
    capture_output=True, text=True
).stdout.strip().split("\n")

# Phase 2: Generate 120 parallel tasks
tasks = []

# 60 backend test agents (one per file)
for i, test_file in enumerate(backend_tests[:60]):
    tasks.append(
        Task(
            subagent_type="general-purpose",
            description=f"QA-BACKEND-{i:02d}",
            prompt=f"""
Run single backend test file:
```bash
cd /Users/.../backend
./venv/bin/python -m pytest {test_file} -v --tb=short 2>&1

Report: PASS/FAIL with test count. """ ) )

50 frontend test agents (one per file)

for i, test_file in enumerate(frontend_tests[:50]): tasks.append( Task( subagent_type="general-purpose", description=f"QA-FRONTEND-{i:02d}", prompt=f""" Run single frontend test file:

cd /Users/.../frontend
npm test -- {test_file} --passWithNoTests 2>&1

Report: PASS/FAIL with test count. """ ) )

10 component batch agents (25 components each)

components = [...254 component files...] batch_size = 25 for i in range(10): batch = components[i*batch_size:(i+1)*batch_size] tasks.append( Task( subagent_type="general-purpose", description=f"QA-COMPONENT-BATCH-{i:02d}", prompt=f""" Check component testability for batch: {chr(10).join(batch)}

For each component, report:

• Has test file: Y/N
• Export issues: Y/N
• Dependencies missing: Y/N """ ) )

Spawn all 120 in parallel

spawn_parallel(tasks)

### Alternative: Pytest-xdist Native Parallelism

If you want pytest's native parallelism instead of Task tool parallelism:

```bash
# Install pytest-xdist
pip install pytest-xdist

# Run with 120 parallel workers
cd backend
pytest tests/ -n 120 --dist loadfile

This uses pytest's built-in parallelism, which:

• Handles test isolation automatically
• Distributes by file (--dist loadfile) or by test (--dist load)
• Aggregates results natively

Hybrid Approach: 12 IG Teams × 10 Test Slices

Model after SEARCH_PARTY structure (using IG instead of G-2):

Each IG team spawns 10 workers = 120 parallel QA agents.

Practical Limits

CRITICAL: IDE Crash Prevention

DO NOT have ORCHESTRATOR spawn 8+ agents directly. This causes IDE seizure and crashes.

CORRECT Pattern:

ORCHESTRATOR → spawns 1 COORD_QUALITY (IG Commander)
                    ↓
              COORD_QUALITY manages 12 IG teams internally
              (uses pytest-xdist, jest parallel, or sequential Task spawns)

WRONG Pattern:

ORCHESTRATOR → spawns 8+ QA agents directly → IDE CRASH

The coordinator absorbs the parallelism complexity. ORCHESTRATOR only ever spawns 1-2 coordinators at a time.

Recommendation

For this codebase:

/qa-party --scale 120

Deploys:

• 12 G-2 QA agents (like SEARCH_PARTY)
• Each G-2 runs 10 test slices via pytest-xdist or jest parallel
• Total: 120 parallel test executions
• Wall-clock: ~60-120 seconds (limited by slowest test suite)

QA_PARTY: Eight agents, one validation, zero marginal cost. The failures are the signal.