Parallel Dispatch

Orchestrate parallel agent execution with dependency awareness.

Research Foundation

• REF-001: BP-9 - Parallel execution for throughput
• REF-002: Archetype 4 - Coordination prevents fragile execution

Usage

/parallel-dispatch tasks.yaml
/parallel-dispatch review-tasks.yaml --max-parallel 4
/parallel-dispatch build-tasks.yaml --timeout 300000 --output .aiwg/working/build/

Task File Format

# tasks.yaml
name: architecture-review
description: Parallel architecture review with synthesis

tasks:
  # Independent tasks run in parallel
  - id: security-review
    agent: security-architect
    prompt: "Review architecture at .aiwg/architecture/sad.md for security"
    output: reviews/security.md
    dependencies: []

  - id: test-review
    agent: test-architect
    prompt: "Review architecture for testability"
    output: reviews/testability.md
    dependencies: []

  - id: ops-review
    agent: devops-engineer
    prompt: "Review architecture for operational concerns"
    output: reviews/operations.md
    dependencies: []

  # Dependent task waits for prerequisites
  - id: synthesis
    agent: documentation-synthesizer
    prompt: "Synthesize all reviews into final assessment"
    output: final-review.md
    dependencies:
      - security-review
      - test-review
      - ops-review

settings:
  max_parallel: 3
  timeout_per_task: 120000
  output_dir: .aiwg/working/reviews/
  on_failure: continue  # or: stop, retry

Options

Option	Default	Description
--max-parallel	4	Maximum concurrent agents
--timeout	120000	Per-task timeout (ms)
--output	.aiwg/working/	Output directory
--on-failure	continue	Failure handling: continue, stop, retry
--dry-run	false	Show execution plan without running

Execution Model

┌─────────────────────────────────────────┐
│           Parallel Dispatch              │
├─────────────────────────────────────────┤
│                                         │
│  Wave 1: Independent Tasks              │
│  ┌───────┐ ┌───────┐ ┌───────┐         │
│  │Task A │ │Task B │ │Task C │         │
│  └───┬───┘ └───┬───┘ └───┬───┘         │
│      │         │         │              │
│      └────────┬──────────┘              │
│               ▼                         │
│  Wave 2: Dependent Tasks                │
│  ┌─────────────────┐                    │
│  │    Task D       │                    │
│  │ (depends on ABC)│                    │
│  └─────────────────┘                    │
│                                         │
└─────────────────────────────────────────┘

Process

1. Parse Tasks: Load task file, validate dependencies
2. Build DAG: Create dependency graph
3. Plan Waves: Group tasks by dependency level
4. Execute Waves: Run each wave in parallel
5. Collect Results: Gather outputs, track failures
6. Report: Summary with timing and status

Output

# Parallel Dispatch Report

## Summary
- **Total Tasks**: 4
- **Succeeded**: 4
- **Failed**: 0
- **Total Time**: 45.2s
- **Parallel Efficiency**: 73%

## Execution Timeline

| Wave | Tasks | Duration | Status |
|------|-------|----------|--------|
| 1 | security, test, ops | 32.1s | ✓ |
| 2 | synthesis | 13.1s | ✓ |

## Task Details

| Task | Agent | Duration | Status | Output |
|------|-------|----------|--------|--------|
| security-review | security-architect | 32.1s | ✓ | reviews/security.md |
| test-review | test-architect | 28.4s | ✓ | reviews/testability.md |
| ops-review | devops-engineer | 25.6s | ✓ | reviews/operations.md |
| synthesis | documentation-synthesizer | 13.1s | ✓ | final-review.md |

## Outputs
- .aiwg/working/reviews/security.md
- .aiwg/working/reviews/testability.md
- .aiwg/working/reviews/operations.md
- .aiwg/working/reviews/final-review.md

Error Handling

on_failure: continue

# Failed tasks don't block independent tasks
# Dependent tasks are skipped if prerequisite fails

on_failure: stop

# First failure stops all execution
# Useful for critical pipelines

on_failure: retry

# Failed tasks retry up to 3 times
# Uses exponential backoff

Examples

Code Review Pipeline

name: code-review-pipeline
tasks:
  - id: security-scan
    agent: security-architect
    prompt: "Security review of PR changes"
    output: security-findings.md

  - id: code-quality
    agent: code-reviewer
    prompt: "Code quality review"
    output: quality-findings.md

  - id: test-coverage
    agent: test-engineer
    prompt: "Test coverage analysis"
    output: coverage-findings.md

  - id: final-review
    agent: code-reviewer
    prompt: "Synthesize all findings into PR review"
    output: pr-review.md
    dependencies: [security-scan, code-quality, test-coverage]

Documentation Generation

name: doc-generation
tasks:
  - id: api-docs
    agent: api-documenter
    prompt: "Generate API documentation"
    output: api/

  - id: user-guide
    agent: technical-writer
    prompt: "Generate user guide"
    output: user-guide/

  - id: architecture-docs
    agent: architecture-documenter
    prompt: "Generate architecture documentation"
    output: architecture/

Related

• prompts/reliability/parallel-hints.md - Parallel patterns
• prompts/core/multi-agent-pattern.md - Multi-agent workflow
• eval-agent --scenario parallel-test - Test parallel execution

Dispatch tasks from: $ARGUMENTS