Metadata-Version: 2.4
Name: openwright-openai-agents
Version: 0.4.0
Summary: OpenAI Agents SDK source connector for OpenWright — capture agent runs, tools, handoffs, and approvals as evidence.
License: Apache-2.0
Keywords: agents,eu-ai-act,evidence,openai-agents,openwright
Requires-Python: <3.15,>=3.10
Requires-Dist: openwright-core<0.8,>=0.6
Provides-Extra: sdk
Requires-Dist: openai-agents>=0.0.1; extra == 'sdk'
Provides-Extra: test
Requires-Dist: openwright-conformance; extra == 'test'
Requires-Dist: pytest>=8; extra == 'test'
Description-Content-Type: text/markdown

# openwright-openai-agents

An **OpenAI Agents SDK source connector** for
[OpenWright](https://github.com/allthingsN/openwright): capture an Agents-SDK
lending agent's runtime (tools, handoffs, guardrail/approval events, the credit
decision) as the same wedge evidence the LangGraph connector produces.

```python
from openwright.connectors import load
conn = load("openwright.source_connectors", "openai_agents")
with conn.instrument(client, thread_id="loan-42") as run:
    # from your tools / hooks:
    run.on_tool_call("credit_check", result)
    appr = run.record_human_approval(reviewer="alice@bank", rationale="reviewed")
    run.record_decision(output="APPROVED", risk_classification="high",
                        approval_ref=appr.event_id, control="art-14-human-oversight")
```

The helper recorders work without the SDK installed; SDK-level auto-capture
attaches (best-effort) when `openai-agents` is present. Payloads are hashed; no
crypto is reimplemented. Not affiliated with OpenAI.
