@zanii/openai-agents

Receipt every tool call an OpenAI Agents SDK run makes — via the SDK's own lifecycle RunHooks, with no changes to your agents. When a tool finishes, its result becomes a signed, verifiable Zanii receipt.

npm install @zanii/openai-agents @zanii/sdk @openai/agents

Usage

import { Runner } from '@openai/agents';
import { ZaniiRunHooks } from '@zanii/openai-agents';
import { ZaniiAgent } from '@zanii/sdk';

const zanii = new ZaniiAgent({ /* … */ });

await Runner.run(agent, input, { hooks: new ZaniiRunHooks(zanii) });
// every completed tool call → a receipt targeting tools.<toolName>

Each receipt targets tools.<toolName> (override with { prefix }) and carries the tool's output. If a SDK version's hook signature differs, call recordToolResult(agent, tool, result) from your own hook — same receipt, wired by hand.

The Agents SDK's tool hooks surface the tool and its result, not the raw input arguments, so receipts record the tool and its output. To capture inputs too, wrap the tool's implementation with the SDK's agent.wrapTool before registering it.

Changelog

• 0.1.0 — initial release: ZaniiRunHooks, zaniiRunHooks, recordToolResult.

License

Apache-2.0.