Agent Protocol

Cognitive Modules compatibility with the Agent Protocol standard.

What is Agent Protocol?

Agent Protocol is an open standard for agent-to-agent communication, enabling interoperability between different AI agent systems.

Compatibility

Cognitive Modules implements a subset of the Agent Protocol specification:

Feature	Status
Task Creation	✅
Task Execution	✅
Task Status	✅
Artifacts	✅
Streaming	🚧

Starting Agent Protocol Server

cog agent-protocol --port 8080

Endpoints

Create Task

POST /ap/v1/agent/tasks

Request:

{
  "input": "Review this code for security issues",
  "additional_input": {
    "module": "code-reviewer",
    "code": "def login(): ..."
  }
}

Response:

{
  "task_id": "task-123",
  "input": "Review this code...",
  "status": "created"
}

Execute Task Step

POST /ap/v1/agent/tasks/:task_id/steps

Response:

{
  "step_id": "step-1",
  "task_id": "task-123",
  "status": "completed",
  "output": {
    "ok": true,
    "meta": {...},
    "data": {...}
  }
}

Get Task

GET /ap/v1/agent/tasks/:task_id

List Artifacts

GET /ap/v1/agent/tasks/:task_id/artifacts

Mapping to Cognitive Modules

Agent Protocol	Cognitive Modules
Task	Module execution request
Step	Single module run
Artifact	Module output
Agent	Module

Example Workflow

import requests

# 1. Create task
task = requests.post(
    'http://localhost:8080/ap/v1/agent/tasks',
    json={
        'input': 'Review code',
        'additional_input': {
            'module': 'code-reviewer',
            'code': '...'
        }
    }
).json()

task_id = task['task_id']

# 2. Execute step
step = requests.post(
    f'http://localhost:8080/ap/v1/agent/tasks/{task_id}/steps'
).json()

# 3. Get result
print(step['output']['data']['issues'])

Multi-Agent Orchestration

Cognitive Modules can participate in multi-agent systems:

# agent-workflow.yaml
agents:
  - name: code-reviewer
    protocol: cognitive
    endpoint: http://cognitive:8000
  
  - name: other-agent
    protocol: agent-protocol
    endpoint: http://other:8080

workflow:
  - agent: code-reviewer
    input: $user_code
  - agent: other-agent
    input: $previous.output