Skip to content

AGENTS.md

Latest commit

 

History

History
84 lines (61 loc) · 2.84 KB

AGENTS.md

File metadata and controls

84 lines (61 loc) · 2.84 KB

AgentMesh — Coding Agent Instructions

Project Overview

AgentMesh is a trust-first communication layer for AI agents — providing cryptographic identity, multi-dimensional trust scoring, scope chains, and governance enforcement for multi-agent systems.

Architecture: 4-layer trust stack

  • Layer 1 (Identity): Ed25519 agent identity, DID (did:mesh:xxx), AI Card integration
  • Layer 2 (Trust): 5-dimension trust scoring, trust decay, handshake protocol, bridges
  • Layer 3 (Governance): Policy enforcement, compliance, hash chain audit chains
  • Layer 4 (Reward): Reputation tracking, reward distribution

Build & Test Commands

# Install dependencies (development mode)
pip install -e ".[dev]"

# Run all tests
pytest tests/ -v

# Run tests with coverage
pytest tests/ --cov=src/agentmesh --cov-report=html

# Type checking
mypy src/

# Lint and format
ruff check .
ruff format .

# Or use Black (also configured)
black --check .

Code Style

  • Formatter: Black (line-length: 100, target: py311)
  • Linter: Ruff (line-length: 100, rules: E, F, I, N, W, UP)
  • Type checker: MyPy strict mode
  • Python: >=3.11
  • Docstrings: Google-style

Key Files

File Purpose
src/agentmesh/identity/ AgentIdentity, AgentDID, Ed25519 key management
src/agentmesh/trust/ TrustScore (5 dimensions), TrustHandshake, TrustBridge
src/agentmesh/governance/ Policy enforcement, compliance checking
src/agentmesh/integrations/ai_card/ AI Card standard adapter (schema, discovery)
src/agentmesh/integrations/a2a/ A2A protocol integration (wraps AICard)
src/agentmesh/services/ Backend services
src/agentmesh/cli/ CLI commands
packages/ External packages (langchain-agentmesh, mcp-proxy)

Coding Conventions

  • All data models use Pydantic BaseModel
  • Identity: AgentIdentity with .sign(data: bytes) -> str (base64 Ed25519 signatures)
  • Trust scores: 5 dimensions — competence, integrity, availability, predictability, transparency
  • DID format: did:mesh:{hex} — derived from public key
  • Private keys stored as _private_key (never serialized)
  • AICard: from_identity() creates signed cards; from_trusted_agent_card() bridges existing formats
  • Tests in tests/ directory

Boundaries

  • Never serialize private keys in JSON/YAML output
  • Never commit secrets, API keys, or credentials
  • Never weaken trust thresholds — only tighten
  • Keep backward compatibility with existing protocol messages
  • packages/ are standalone — changes there need their own test suite

Testing Requirements

  • All new features must include tests
  • Run pytest tests/ -v before committing
  • Use pytest-asyncio for async tests (asyncio_mode = "auto")

Commit Style

Use conventional commits: feat:, fix:, docs:, test:, refactor:, chore: