TrustProof-Protocol

OpenAI Agents Integration

What It Does

createTrustProofAgentHook(...) provides SDK-agnostic hook handlers for agent lifecycle events.

For each completed action/tool event, it:

• computes canonical input/output hashes
• creates a claims envelope
• signs a JWT proof
• chains proofs in memory when enabled

Implementation:

• ../../packages/js/src/integrations/openai_agents.ts

Run the Example

pnpm --filter @trustproof/sdk example:openai-agents

Example script:

• ../../examples/integrations/openai_agents/on_action_proof_chain.mjs

Minimal Usage

import { createTrustProofAgentHook } from "@trustproof/sdk";

const hook = createTrustProofAgentHook({
  privateKeyPem,
  subject: { type: "agent", id: "agent_demo_1" },
  policy: { policy_v: "v0", scopes: ["tools:invoke"], constraints: {} },
  chain: { enabled: true }
});

const { proof_jwt } = await hook.onToolEnd({
  tool_name: "payout.initiate",
  input,
  output,
  action_name: "agent.tool.payout.initiate"
});

Connecting to Real Agents SDK Callbacks

Keep the adapter thin:

• map SDK callback payloads into the internal event shape (action_name, tool_name, input, output, resource, timestamp, jti)
• call onToolEnd(...) / onActionEnd(...) at completion points
• persist getChain().proofs externally if you need cross-process continuity

Avoid deep coupling to transient SDK types; keep mapping logic isolated.

This site is open source. Improve this page.