[P0] Sprint 2.1b: AutoGen Trinity Consensus (Detailed)

[ariffazil]

Objective

Replace L5 agent stubs with AutoGen GroupChat for Tri-Witness consensus.

Repository

https://github.com/microsoft/autogen

Effort

5 days

Implementation

Trinity Agents

from autogen import AssistantAgent, UserProxyAgent, GroupChat

# AGI Mind
architect = AssistantAgent(
    "agi_architect",
    system_message="You are AGI Mind. Question: Is it TRUE? Enforce F2, F4, F7.",
    llm_config={"model": "gpt-4"}
)

# ASI Heart
guardian = AssistantAgent(
    "asi_guardian",
    system_message="You are ASI Heart. Question: Is it SAFE? Enforce F5, F6, F9.",
    llm_config={"model": "gpt-4"}
)

# APEX Soul (F13)
sovereign = UserProxyAgent(
    "apex_sovereign",
    human_input_mode="ALWAYS",
    code_execution_config=False
)

# Tri-Witness GroupChat
trinity = GroupChat(
    agents=[architect, guardian, sovereign],
    messages=[],
    max_round=3
)

Integration

@mcp.tool()
async def apex_verdict(query: str, session_id: str) -> dict:
    result = trinity.run(query)
    w3 = calculate_tri_witness(result)
    return {"verdict": "SEAL" if w3 >= 0.95 else "SABAR"}

Subtasks

• Install AutoGen: pip install pyautogen
• Create codebase/agents/trinity.py
• Implement 3-agent debate
• Calculate W₃ consensus
• Human approval (F13) via CLI/web
• Test Tri-Witness calculation
• Update L5_AGENTS status

Acceptance Criteria

• 3-agent debate runs (AGI → ASI → Human)
• W₃ ≥ 0.95 enforcement works
• Human can veto (F13)
• L5_AGENTS status changes from 🔴 to 🟡

Files Changed

• codebase/agents/trinity.py (new)
• codebase/mcp/tools/apex_tool.py
• 333_APPS/L5_AGENTS/STATUS.md

Related

• Issue [P0] Sprint 2.1: Integrate AutoGen for Multi-Agent Debate #167
• INTEGRATION_MASTERPLAN.md

[github-actions]

🤖 Hi @ariffazil, I've received your request, and I'm working on it now! You can track my progress in the logs for more details.

[github-actions]

🤖 I'm sorry @ariffazil, but I was unable to process your request. Please see the logs for more details.