🧠 Debate & Memory System

MOP implements a high-performance AI Debate Engine that automatically triggers before any critical execution step, using five specialized agents to achieve consensus through structured competition.

1. Debate Participants (Debate Mode)

Agent	Core Objective	Primary Focus
Analyst Agent	Problem structuring	Risk identification and scoping options
Solution Agent	Optimal blueprint	Efficiency and swift implementation
Critic Agent	Destructive analysis	Finding hidden bugs and edge-case errors
Security Agent	Threat evaluation	Checking file system bounds, secret containment
Optimizer Agent	Refinement	Proposing alternative paths and performance tuning

2. Debate Workflow

The debate process follows a strict 6-stage sequence:

[STEP 1] Analyst Agent establishes the initial context and problem structure.
    │
[STEP 2] Solution Agent proposes the optimal blueprint.
    │
[STEP 3] Critic Agent challenges the solution, hunting for structural flaws.
    │
[STEP 4] Security Agent evaluates security constraints and credential hygiene.
    │
[STEP 5] Optimizer Agent offers alternative paths and optimizations.
    │
[STEP 6] The Aggregation Engine weighs all positions and resolves the consensus.

---

3. Debate Aggregation Engine

The Aggregation Engine collects all arguments and calculates a Confidence Score for each agent:

• Compares arguments and computes position strength scores.
• Integrates Memory-Based Weighting (if a Solution Agent historically proposed highly stable code, its current blueprint score is weighted higher; if the Critic historically predicted a bug, its current objection is weighted heavier).
• Identifies contradictions and guarantees a safe fallback resolve.

Every decision resolved by this engine is completely explainable, traceable, and reproducible.

---

4. SQLite Persistent Memory

The Memory System is integrated directly into the debate cycle. The SQLite database consists of:

• debates - Session metadata, timestamps, trace IDs, and final consensus decisions.
• arguments - Individual arguments posted by agents during active debate sessions.
• decision_history - Comprehensive historical log of all finalized actions.
• agent_opinion_history - Agent opinion records, tracking agent alignment trends over time.

Similarity Lookup

Before a debate begins, the system queries the SQLite database for similar past workflows. If a similar past approach resulted in a failed post-flight test or a rollback, the Critic Agent is automatically injected with this historical context. The Critic can then mount a highly targeted argument to prevent repeating the same mistake.