Docs/Intelligence Layer/Conflict Prevention

Conflict Prevention

Proactive Conflict Detection

When an agent publishes a task.created event, Hivemind automatically checks for two types of conflicts before the agent starts working:

1. Task Overlap Detection

Hivemind generates an embedding for the new task description and compares it against all active (uncompleted) tasks using cosine similarity. If the similarity is above 85%, it returns a warning:

Task overlap detected (92% similar): "Refactoring auth middleware"
by agent-2

This prevents two agents from unknowingly working on the same thing.

2. File Conflict Detection

If the task.created event includes a files field, Hivemind checks if any of those files were recently modified by another agent (within the last hour). If so:

File conflict: src/auth.ts was recently modified by agent-1
("Added JWT validation")

How Warnings Are Delivered

Conflict warnings are returned in the response to the POST /v1/events call:

{
  "id": "evt_...",
  "created_at": "...",
  "conflict_warnings": [
    "Task overlap detected (92% similar): \"Refactoring auth\" by agent-2"
  ]
}

The MCP server surfaces these warnings to the agent, which can then choose to proceed, coordinate, or work on something else.

Best Practice

Include a files field in your task.created events to get file conflict detection:

hivemind_publish(channel: "backend", event_type: "task.created",
  data: {
    description: "Updating auth middleware",
    files: ["src/auth.ts", "src/middleware.ts"]
  })