dispatching-parallel-agents

# Dispatching Parallel Agents (Shannon-Enhanced)

## Overview

**Parallel investigation of independent failures with quantitative success tracking.**

Dispatch one agent per independent problem domain. Shannon enhancement adds wave-based coordination, numerical success scoring, and MCP result aggregation.

**Success Scoring (0.00-1.00):**
- Per-domain: agent_success_score = (problems_fixed / problems_identified)
- Overall: parallel_efficiency = (domains_completed_concurrently / sequential_cost)

## When to Use

```
3+ independent failures?
├─ Yes, independent domains?
│  ├─ Yes → Parallel dispatch (optimal)
│  └─ No → Sequential investigation
└─ No → Single agent focus
```

**Dispatch when:**
- 3+ test files failing with different root causes
- Multiple subsystems broken independently
- Each problem understood without context from others
- No shared state between investigations

**Don't dispatch:**
- Failures are related (fix one might fix others)
- Agents need full system state understanding
- Would interfere with each other's work

## The Pattern

### 1. Identify Independent Domains

Group failures by subsystem:
```
Domain A: Tool approval flow (file_a_test)
Domain B: Batch completion (file_b_test)
Domain C: Abort functionality (file_c_test)
```

**Independence check:** Fixing Domain A doesn't touch Domain B/C code paths.

### 2. Assign Success Metrics

```bash
# Per-domain scoring (Serena)
per_domain_metrics = {
  domain: "Tool approval",
  problems_identified: 3,
  initial_root_cause_clarity: 0.6,  # 0-1.0
  estimated_complexity: 0.7,        # 0-1.0
}
```

### 3. Dispatch with Shannon Wave Coordination

```bash
# Launch agents concurrently with wave tracking
Task("Fix Domain A", wave_id="w1", timeout=30min)
Task("Fix Domain B", wave_id="w1", timeout=30min)
Task("Fix Domain C", wave_id="w1", timeout=30min)

# Shannon wave monitors parallel execution
# MCP tracks: start_time, end_time, status per agent
```

### 4. Aggregate Results via MCP

**Result structure (Ser