Skip to content

wharfe/agent-trust-suite

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent Trust Suite

Trust infrastructure for multi-agent systems. Define contracts. Observe behavior. Verify traces.

As multi-agent systems grow through protocols like A2A and MCP, interoperability is advancing rapidly — but the trust layer remains thin. Agents can talk to each other, but there is no standard way to verify that they behave as expected.

Agent Trust Suite is an open-source collection of tools that makes agent behavior observable and verifiable across the execution lifecycle.

Agent Trust Suite Demo — contract → observation → evidence

What This Suite Helps You Verify

  • That agents act within pre-agreed contracts
  • That deviations are detected in real time during execution
  • That verifiable evidence trails remain after execution

3 Layers

┌─────────────────────────────────────────────────────────┐
│  Before Execution                                       │
│  agentcontract — Define expected behavior as contracts  │
├─────────────────────────────────────────────────────────┤
│  During Execution                                       │
│  agent-trust-telemetry — Observe trust risks in real    │
│  time                                                   │
├─────────────────────────────────────────────────────────┤
│  After Execution                                        │
│  trustbundle — Preserve tamper-evident evidence trails   │
└─────────────────────────────────────────────────────────┘
        ▲
        │  Authorization Substrate
        │
  agentbond — Shared governance infrastructure
  powering authorization, intent tracking, and contracts
  across all three layers (MCP Server / SDK)

Quick Start

Try the Demo

See the full trust lifecycle in 60 seconds — no API keys required:

npx agent-trust-cli demo

Or use individual commands:

npx agent-trust-cli verify bundle.json       # Verify a trust bundle
npx agent-trust-cli inspect contract.yaml    # Inspect any trust artifact

For live telemetry evaluation in the demo, also install: pip install agent-trust-telemetry

Individual Tools

Each tool can be used independently. Pick the layer relevant to your use case:

agentcontract (Node.js) — Define and validate agent behavior contracts

npm install -g agentcontract
agentcontract init --name my-agent
agentcontract run my-agent.contract.yaml

agent-trust-telemetry (Python) — Detect instruction contamination across agent traces

pip install agent-trust-telemetry          # if published
# or install from source:
git clone https://github.com/wharfe/agent-trust-telemetry.git
cd agent-trust-telemetry && pip install -e ".[dev]"

att evaluate --message message.json
att evaluate --stream trace.jsonl

trustbundle (Node.js) — Package execution traces into tamper-evident bundles

npm install -g trustbundle               # if published
# or install from source:
git clone https://github.com/wharfe/trustbundle.git
cd trustbundle && npm install && npm run build

trustbundle init
trustbundle build <trace.jsonl>
trustbundle verify <bundle.json>

agentbond (Node.js monorepo) — Governance infrastructure with MCP Server

npx @agentbond/mcp-server                # run MCP server directly
# or use as a library:
npm install @agentbond/auth @agentbond/intent @agentbond/contract @agentbond/settlement

Unified CLI

agent-trust-cli provides a single entry point with demo, verify, and inspect commands. CLI name may change in future releases.

Architecture

See docs/architecture.md for the detailed data flow between components.

graph TB
    subgraph "Before"
        AC[agentcontract<br/>Contract Definition & Validation]
    end
    subgraph "During"
        ATT[agent-trust-telemetry<br/>Runtime Trust Observation]
    end
    subgraph "After"
        TB[trustbundle<br/>Evidence Packaging]
    end
    subgraph "Substrate"
        AB[agentbond<br/>Authorization & Governance]
    end

    AC -->|"contract scope"| ATT
    ATT -->|"trace data"| TB
    AB -.->|"auth tokens<br/>intent records<br/>contract lifecycle"| AC
    AB -.->|"authorization decisions"| ATT
    AB -.->|"audit records"| TB
Loading

Status

Repository Version Stage Language
agentcontract 0.1.0 MVP — API may change Node.js
agent-trust-telemetry 0.1.0 Alpha — MVP phase complete Python
trustbundle 0.1.0 MVP — digest-based integrity Node.js
agentbond 0.1.0 MVP — 17 MCP tools available Node.js

All components are in early development. APIs and interfaces may change.

Contributing

Contributions are welcome! Each repository accepts issues and pull requests independently.

For suite-level discussions (cross-repo architecture, new layer proposals, demo scenarios), please use this repository's Issues.

License

MIT

About

Trust infrastructure for multi-agent systems. Define contracts. Observe behavior. Verify traces.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors