max-subagents-parallel

name	max-subagents-parallel
description	Use when starting ANY non-trivial task to maximize parallelism - spawns multiple subagents, creates task graphs, identifies independent work streams

Maximum Parallelism Mode

Overview

Default to parallel, not sequential. Before executing ANY multi-step task, first map dependencies then spawn maximum concurrent agents (target: 8-10 per wave).

Strategies

Three execution strategies available (default: recursive):

Strategy	Best For	Description
recursive	General tasks	Decompose → Execute in waves → Synthesize
ensemble	High-stakes decisions	Same subtask → 2-3 models independently → merge/compare
sweep	Codebase understanding	Max breadth Wave 1 (up to 10 agents), each a different angle

The Pattern

digraph max_parallel {
    rankdir=TB;

    "Receive task" [shape=box];
    "Select strategy (recursive/ensemble/sweep)" [shape=box];
    "Map all subtasks (target 8-10)" [shape=box];
    "Identify dependencies (blocks/blockedBy)" [shape=box];
    "Group into parallel waves" [shape=box];
    "Spawn agents for Wave 1 (up to 10 independent tasks)" [shape=box];
    "Wait for Wave 1 completion" [shape=diamond];
    "Spawn Wave 2 (tasks blocked by Wave 1)" [shape=box];
    "Continue until all waves complete" [shape=box];
    "Synthesis Wave - merge & validate all results" [shape=box, style=bold];

    "Receive task" -> "Select strategy (recursive/ensemble/sweep)";
    "Select strategy (recursive/ensemble/sweep)" -> "Map all subtasks (target 8-10)";
    "Map all subtasks (target 8-10)" -> "Identify dependencies (blocks/blockedBy)";
    "Identify dependencies (blocks/blockedBy)" -> "Group into parallel waves";
    "Group into parallel waves" -> "Spawn agents for Wave 1 (up to 10 independent tasks)";
    "Spawn agents for Wave 1 (up to 10 independent tasks)" -> "Wait for Wave 1 completion";
    "Wait for Wave 1 completion" -> "Spawn Wave 2 (tasks blocked by Wave 1)" [label="done"];
    "Spawn Wave 2 (tasks blocked by Wave 1)" -> "Continue until all waves complete";
    "Continue until all waves complete" -> "Synthesis Wave - merge & validate all results";
}

Execution Rules

1. ALWAYS Map Before Acting

Before ANY implementation:

1. List ALL subtasks — target 8-10 per wave (don't stop at 2-3)
2. For each pair: "Does A need B's output?" → B blocks A
3. Tasks with no blockers = Wave 1 (parallel)
4. Tasks blocked only by Wave 1 = Wave 2 (parallel after Wave 1)
5. After all waves: Synthesis wave (MANDATORY)

2. ALWAYS Use Parallel Tool Calls

WRONG - Sequential:

Task("Explore auth module")
[wait for result]
Task("Explore database module")
[wait for result]
Task("Explore API module")

RIGHT - Parallel (single message, multiple tool calls):

Task("Explore auth module")     // These all spawn
Task("Explore database module") // in a single message
Task("Explore API module")      // executing concurrently
Task("Explore state mgmt")     // Keep going —
Task("Explore config layer")   // target 8-10 agents
Task("Explore error handling") // per wave

3. Match Model to Task Complexity

Task Type	Model	Why
File exploration, glob, grep	haiku	Fast, cheap, sufficient
Code reading, understanding	sonnet	Balance of speed/capability
Architecture, complex logic	opus	Deep reasoning needed
Ensemble verification	mix all 3	Different perspectives
Synthesis / final merge	opus	Needs to reconcile all findings

4. Use TaskCreate with Dependencies

TaskCreate("Implement auth service")     // id: 1
TaskCreate("Implement user model")       // id: 2
TaskCreate("Implement API routes")       // id: 3, blocks: [1, 2]
TaskCreate("Write integration tests")    // id: 4, blockedBy: [3]

Tasks 1 and 2 run parallel. Task 3 waits for both. Task 4 waits for 3.

5. Mandatory Synthesis Wave

After all execution waves complete, ALWAYS spawn a synthesis agent:

Task("Synthesize and validate all results from Waves 1-N", subagent_type="general-purpose", model="opus")

The synthesis agent must:

• Merge findings from all agents into a coherent summary
• Flag contradictions or conflicts between agent results
• Validate completeness — did we miss anything?
• Produce the final deliverable

Quick Reference

Scenario	Action
Exploring codebase	Spawn 8-10 Explore agents with different focuses simultaneously
Multiple test failures	One agent per test file in parallel (up to 10)
Feature implementation	Map all components, spawn parallel for independent ones
Research questions	Spawn multiple search agents with different queries
Code review	Parallel agents for different aspects (security, performance, style, correctness)
High-stakes decision	Use ensemble strategy: same question → 3 models → merge
Open-ended exploration	Use sweep strategy: 10 agents, each a different angle

Example: Feature Implementation (Recursive Strategy)

Task: "Add user authentication"

WRONG (sequential):

1. Research auth patterns [wait]
2. Implement user model [wait]
3. Implement auth service [wait]
4. Add API routes [wait]
5. Write tests [wait]

RIGHT (parallel waves, 10 agents):

Wave 1 (parallel - no dependencies, 6 agents):
  - Task("Research auth patterns in codebase", model="haiku")
  - Task("Research existing user model code", model="haiku")
  - Task("Research API route patterns", model="haiku")
  - Task("Research middleware patterns", model="haiku")
  - Task("Research test patterns", model="haiku")
  - Task("Research config/env for auth secrets", model="haiku")

Wave 2 (parallel - after Wave 1, 3 agents):
  - Task("Implement user model", model="sonnet")
  - Task("Implement auth service", model="sonnet")
  - Task("Implement auth middleware", model="sonnet")

Wave 3 (after Wave 2):
  - Task("Add API routes", blockedBy=[user_model, auth_service, middleware])

Wave 4 (after Wave 3):
  - Task("Write tests", blockedBy=[api_routes])

Wave 5 (Synthesis - MANDATORY):
  - Task("Validate auth implementation: review all changes, run tests, check security", model="opus")

Result: 5 waves with up to 6 agents in Wave 1. Massively faster than sequential.

Example: Ensemble Strategy

Task: "Should we use Redis or PostgreSQL for session storage?"

Wave 1 (ensemble - same question, 3 models):
  - Task("Analyze: Redis vs PostgreSQL for session storage given our codebase", model="haiku")
  - Task("Analyze: Redis vs PostgreSQL for session storage given our codebase", model="sonnet")
  - Task("Analyze: Redis vs PostgreSQL for session storage given our codebase", model="opus")

Wave 2 (synthesis):
  - Task("Compare 3 independent analyses, flag agreements/disagreements, produce recommendation", model="opus")

Example: Sweep Strategy

Task: "Understand this new codebase"

Wave 1 (sweep - 10 agents, maximum breadth):
  - Task("Explore: project structure and entry points", model="haiku")
  - Task("Explore: database models and schema", model="haiku")
  - Task("Explore: API routes and controllers", model="haiku")
  - Task("Explore: authentication and authorization", model="haiku")
  - Task("Explore: frontend components and state", model="haiku")
  - Task("Explore: test infrastructure and coverage", model="haiku")
  - Task("Explore: build system and CI/CD", model="haiku")
  - Task("Explore: configuration and environment", model="haiku")
  - Task("Explore: error handling and logging", model="haiku")
  - Task("Explore: third-party integrations", model="haiku")

Wave 2 (synthesis):
  - Task("Synthesize all 10 explorations into architecture overview", model="opus")

Trigger Phrases

When user says any of these, invoke /max-subagents-parallel:

• "do this in parallel"
• "spawn multiple agents"
• "maximize parallelism"
• "work faster"
• "parallelize this"

Anti-Patterns (NEVER DO)

• Starting with step 1 before mapping all steps
• Sequential tool calls when parallel is possible
• Using Opus for file exploration
• Single-threaded investigation of independent failures
• Waiting for one search before starting another unrelated search
• Stopping at 3-5 agents when 8-10 would be more effective
• Skipping the synthesis wave

The Iron Law

If tasks are independent, they MUST run in parallel.
Sequential execution of independent tasks is a bug.