Files
manual_slop/conductor/tracks/mma_verification_20260225/spec.md

2.2 KiB

Specification: MMA Tiered Architecture Verification

Overview

This track aims to review and verify the implementation of the 4-Tier Hierarchical Multi-Model Architecture (MMA) within the Conductor framework. It will confirm that Conductor operates as a Tier 2 Tech Lead/Orchestrator and can successfully delegate tasks to Tier 3 (Workers) and Tier 4 (QA/Utility) sub-agents. A key part of this track is investigating whether this hierarchy should be enforced via a single centralized skill or through separate role-based sub-agent definitions.

Functional Requirements

  1. Skill Review: Analyze mma-orchestrator/SKILL.md and MMA_Support docs to ensure they correctly mandate Tier 2 behavior for Conductor.
  2. Delegation Verification:
    • Verify Conductor (Tier 2) can spawn Tier 3 sub-agents for heavy coding tasks using .\scripts un_subagent.ps1.
    • Verify Tier 3/4 sub-agents can be spawned for error analysis/compression.
  3. Architectural Investigation: Evaluate the pros/cons of a centralized mma-orchestrator skill vs. independent role-based sub-agents. Determine the best way to define sub-agent roles.
  4. Test Track Creation: Implement a "Mock Implementation" track that demonstrates the full tiered delegation flow (Tier 2 -> Tier 3 -> Tier 4).
  5. Automated Testing: Create pytest cases to verify the IPC and script execution flow of the tiered sub-agents.

Non-Functional Requirements

  • Traceability: All sub-agent invocations must be clearly logged in the session.
  • Context Efficiency: Ensure sub-agent delegation effectively prevents token bloat in the main Conductor context.

Acceptance Criteria

  • Analysis report comparing centralized skill vs. role-based sub-agents.
  • A functional test track (mma_verification_mock) that executes a full tiered delegation sequence.
  • Traceable logs confirming sub-agent spawning and task completion.
  • Pytest suite verifying the sub-agent infrastructure and interaction logic.
  • Plan alignment: The test track's plan.md explicitly includes delegation steps.

Out of Scope

  • Implementing a full production-ready multi-model backend.
  • Modifying the core gemini-cli binary.