parallel-dispatch

parallel-dispatch

Generic parallel agent orchestration utility for launching multiple agents concurrently.

Triggers

Alternate expressions and non-obvious activations (primary phrases are matched automatically from the skill description):

• "fan out" → parallel agent dispatch shorthand
• "run all [N] in parallel" → explicit parallel execution

Purpose

This skill provides the foundational infrastructure for launching multiple agents in parallel, collecting their results, and handling timeouts. It is used by higher-level skills like artifact-orchestration, review-synthesis, and gate-evaluation.

Behavior

When triggered, this skill:

1. Parses agent configuration:

   • Agent names (from built-in or custom)
   • Prompt templates for each agent
   • Shared context/artifact reference
   • Timeout settings

2. Prepares agent-specific prompts:

   • Loads prompt template per agent
   • Injects shared context (artifact path, requirements)
   • Adds output format requirements

3. Launches agents in parallel:

   • Uses single message with multiple Task tool calls
   • Each agent runs independently
   • No inter-agent communication during execution

4. Collects results:

   • Waits for all agents or timeout
   • Captures success/failure per agent
   • Structures results for downstream processing

5. Returns consolidated results:

   • Per-agent output
   • Execution metadata (duration, status)
   • Aggregated insights (if configured)

Configuration Format

dispatch:
  name: "artifact-review"
  timeout: 300  # seconds

  agents:
    - name: security-architect
      prompt: |
        Review the artifact at {artifact_path} for security concerns.
        Focus on: authentication, authorization, data protection, input validation.
        Output format: structured findings with severity ratings.

    - name: test-architect
      prompt: |
        Review the artifact at {artifact_path} for testability.
        Focus on: test coverage gaps, edge cases, integration points.
        Output format: test recommendations with priority.

    - name: requirements-analyst
      prompt: |
        Review the artifact at {artifact_path} for requirements traceability.
        Focus on: requirement coverage, gaps, conflicts.
        Output format: traceability assessment.

  context:
    artifact_path: ".aiwg/architecture/sad.md"
    requirements_path: ".aiwg/requirements/"

  result_format: structured  # or 'raw'
  aggregate: true  # combine findings