Swarm Decomposition

Swarm Decomposition (DAG Workflow)

Complex goals are decomposed into subtasks using Grok, then executed as a Directed Acyclic Graph (DAG) by multiple agents.

Strategies

Parallel

All subtasks created with no dependencies. Every agent can claim immediately.

{ "goal": "...", "strategy": "parallel", "agent_count": 4 }

Sequential

Each subtask depends on the previous one. Forms a linear chain.

{ "goal": "...", "strategy": "sequential", "agent_count": 1 }

DAG (Directed Acyclic Graph)

Grok analyzes the goal and creates subtasks with explicit dependency links. Tasks without dependencies start immediately; blocked tasks wait.

{ "goal": "...", "strategy": "dag", "agent_count": 3 }

Dependency Resolution

When a task completes:

1. Orchestrator queries all tasks with depends_on containing the completed task ID
2. For each dependent task, checks if all its dependencies are Completed
3. If yes and an agent is pre-assigned, auto-transitions to Claimed
4. If yes and no agent assigned, transitions to Open for discovery

Example: Research Swarm

Goal: "Research OPTX competitors, analyze their tokenomics, and draft a comparison report"

Grok decomposes into:

• Task A: "Research competitor projects on Solana" (no deps)
• Task B: "Analyze DePIN token models" (no deps)
• Task C: "Draft comparison report" (depends on A, B)

Tasks A and B execute in parallel. Task C waits until both complete, then auto-starts.

Related

• Task Lifecycle — Single task flow
• Task State Machine — All possible states