Hive Mind Protocol: Swarm Intelligence

This skill defines the architecture for "Enxame de Agentes" (Agent Swarms) where multiple specialized instances collaborate to solve distributed problems (e.g., scraping multiple sites simultaneously).

1. The Swarm Concept

Instead of one agent doing everything, we deploy a "swarm" of identical agents configured with different parameters (e.g., differents URLs or datasets).

1.1 Dynamic Agent Creation

Iterate through a list of targets to spawn agents dynamically.

python

agents = []
tasks = []
targets = ["site1.com", "site2.com", "site3.com"]

for idx, target in enumerate(targets):
    # Specialized Agent
    agent = Agent(
        role=f"Scout Unit {idx}",
        goal=f"Extract intelligence from {target}",
        backstory="You are a specialized scout...",
        tools=[WebsiteSearchTool(website=target)],
        verbose=True
    )
    
    # Specialized Task
    task = Task(
        description=f"Analyze {target} and report findings.",
        expected_output="Key intelligence points.",
        agent=agent,
        output_file=f"intel_{idx}.md"
    )
    
    agents.append(agent)
    tasks.append(task)

2. The Aggregator

A Swarm must always resolve to a single point of truth. The Aggregator Agent collects the outputs of the swarm.

python

aggregator = Agent(
    role="Mind Nexus",
    goal="Synthesize all scout reports into a master strategy.",
    backstory="You are the central processing unit..."
)

aggregator_task = Task(
    description="Read all field reports and create a summary.",
    expected_output="Master Strategy Document (Markdown).",
    agent=aggregator,
    context=tasks # Pass previous tasks as context
)

# Add to the end of the lists
agents.append(aggregator)
tasks.append(aggregator_task)

3. Execution Strategy

Run the swarm sequentially or asynchronously depending on resource limits.

python

hive_crew = Crew(
    agents=agents,
    tasks=tasks,
    process=Process.sequential # or Process.hierarchical for complex management
)

result = hive_crew.kickoff()

4. Short-Term Memory

Enable memory=True in agents to allow them to recall previous steps and avoid loops during complex navigations.

5. Protocol Rules

•Isolation: Swarm agents should not depend on each other, only on their specific target.
•Synthesis: The final output must ALWAYS be aggregated. A defined "Swarm" without a "queen" (aggregator) is just noise.
•Error Handling: Use try/except when spawning agents for potentially unstable targets (e.g., broken URLs) to prevent the entire swarm from failing initialization.