Skip to content

openai-agent-transcript-example.md

Latest commit

 

History

History
87 lines (68 loc) · 2.55 KB

openai-agent-transcript-example.md

File metadata and controls

87 lines (68 loc) · 2.55 KB

OpenAI Agent Transcript Example

This is an expected transcript shape for the OpenAI Agents SDK adapter. It is not a captured live OpenAI trace; it documents the interaction pattern that a live trace should show once openai-agents and OPENAI_API_KEY are configured.

Case

case-002: ambiguous request requiring clarification.

Inbound message:

From: Ben Carter <ben.carter@example.test>
Subject: Sales for last quarter

Hi,

Can I get sales for last quarter by region? Need it for a Monday deck.

Thanks,
Ben

Expected Agent/Tool Flow

User input
  -> Intake agent reads email text
  -> Intake agent extracts requester email: ben.carter@example.test
  -> Intake agent extracts partial structured request
  -> Intake agent calls process_sales_report_request
  -> Tool resolves requester profile: u-1002
  -> Portable core validates request
  -> Portable core returns clarification_required
  -> Intake agent returns structured final output

Expected Tool Input

{
  "request_id": "case-002",
  "requester_email": "ben.carter@example.test",
  "structured_request": {
    "report_type": "sales_summary",
    "metrics": [],
    "ambiguous_terms": {
      "sales": ["revenue", "units", "gross_margin"]
    },
    "dimensions": ["region"],
    "date_range": {
      "start": "2026-01-01",
      "end": "2026-03-31",
      "label": "last quarter",
      "reference_date": "2026-05-05"
    },
    "filters": {},
    "output_format": null,
    "purpose": "Monday deck"
  }
}

Expected Final Output

{
  "request_id": "case-002",
  "requester_id": "u-1002",
  "outcome": "clarification_required",
  "response_message": "I can continue, but need clarification before generating a report. Please reply in this email thread and confirm whether sales means revenue, units, gross margin, or all permitted metrics; and confirm the country or region for the report.",
  "output_path": null
}

What A Live SDK Trace Should Show

The OpenAI Agents SDK trace should show:

  • one Runner run for the request
  • one intake-agent turn
  • one function-tool call to process_sales_report_request
  • no report-generation tool inside the model layer
  • final structured output matching the tool result

This confirms the intended boundary: the model interprets; the portable core validates, decides, audits, and generates outputs.

For Together AI runs, OpenAI-hosted tracing is disabled by default. The same expected flow still applies; use the returned tool result, generated audit event, and local terminal output as the transcript unless tracing is explicitly enabled.