Agent Contracts Reference

Quick Reference

• Interaction Pattern: Async JSON-RPC
• Auth: Local Environment / API Keys
• Primary Format: Structured JSON

Overview

This documentation defines the input and output contracts for the specialized agents within the orchestrator. Adhering to these contracts ensures deterministic execution and reliable parsing.

Available Agent Roles

Agent	Purpose	Primary Interface
Analyst	Workflow Decomposition	execute(request) ➜ DAG JSON
Solution	Technical Implementation	execute(context) ➜ CODE JSON
Critic	Peer Review	execute(proposal) ➜ AUDIT JSON
Security	Vulnerability Scan	execute(proposal) ➜ RISK JSON

---

Analyst Agent Contract

Generates a Directed Acyclic Graph (DAG) for implementation.

Request Body

{
  "request": "Implement a logging system."
}

Response Body

200 OK (Validated Plan)

{
  "tasks": [
    {
      "id": 1,
      "agent": "CodingAgent",
      "description": "Initialize logger configuration.",
      "params": { "query": "..." },
      "depends_on": []
    }
  ],
  "risks": ["Performance overhead"]
}

Examples

Python (Direct Call)

from agents.analysis_agent import AnalysisAgent

agent = AnalysisAgent(memory_manager, logger)
plan = await agent.execute({"request": "Task..."}, trace_id="tr_123")

---

Model Router API

Handles model failover and retries.

Method	Description	Fallback
execute_with_routing	Runs a completion with auto-retry.	Gemini ➜ Mistral

Error Codes

Code	Name	Description
429	Rate Limit	Primary model quota exhausted.
500	Context Failed	All models in fallback chain failed.

---

See Also

• System Architecture
• Database Schema