Skip to content

METAGPT-INTEGRATION-PROPOSAL.md

Latest commit

 

History

History
82 lines (60 loc) · 2.68 KB

METAGPT-INTEGRATION-PROPOSAL.md

File metadata and controls

82 lines (60 loc) · 2.68 KB

MetaGPT — Trust Layer Integration

Submission: FoundationAgents/MetaGPT#1936 Status: Open PR — awaiting review Type: Extension module (metagpt.ext.agentmesh) Date Submitted: March 2, 2026

Summary

Inter-agent trust verification for MetaGPT multi-agent teams using Agent-Mesh. Adds cryptographic identity (DIDs), trust scoring, capability-based access control, and audit logging to MetaGPT's role-based architecture.

Problem

MetaGPT enables powerful multi-agent collaboration through specialized roles, but agents interact without verifying trust:

  • ProductManager sends requirements → Is PM authorized?
  • Architect designs system → Can we trust the design?
  • Engineer writes code → Should we execute untrusted code?

Without trust verification, a compromised role in the pipeline can inject malicious instructions that propagate through the entire team.

Solution

TrustedTeam

from metagpt.ext.agentmesh import TrustedTeam, TrustPolicy, TrustLevel

policy = TrustPolicy(
    min_trust_level=TrustLevel.MEDIUM,
    sensitive_actions={"WriteCode", "ExecuteCode"},
    sensitive_action_trust=TrustLevel.HIGH,
)

team = TrustedTeam(policy=policy)
team.add_role(ProductManager(), trust_level=TrustLevel.HIGH)
team.add_role(Engineer(), trust_level=TrustLevel.MEDIUM)

# Verifies trust before interaction
team.verify_message("ProductManager", "Engineer", "AssignTask")

Key Components

Component Purpose
TrustedRole Wraps MetaGPT roles with cryptographic identity
TrustPolicy Configurable trust requirements per team
TrustVerifier Verifies interactions between agents
TrustedTeam Team wrapper with enforcement

Trust Levels

Level Score Use Case
LOW 0.0–0.3 Read-only roles, observers
MEDIUM 0.3–0.7 Standard collaboration roles
HIGH 0.7–1.0 Code execution, system modifications

Value Proposition

Feature Without Trust With Agent-Mesh
Agent Identity None Cryptographic DIDs
Interaction Control None Policy-based
Sensitive Actions Unrestricted Trust-gated
Audit Trail None Full logging

Files Added

metagpt/ext/agentmesh/
├── __init__.py      # Public exports
├── trust_layer.py   # Core trust primitives
└── README.md        # Documentation

Links