Description

When using openai/gpt-5.2-chat-latest as the primary model, after a few turns of conversation, both WhatsApp and Telegram channels return:

400 Item 'rs_...' of type 'reasoning' was provided without its required following item.

The error recurs every time the session accumulates reasoning items from GPT-5.2's responses. Clearing the session temporarily fixes it, but after a few more messages the error comes back.

Steps to Reproduce

1. Configure agents.defaults.model.primary to openai/gpt-5.2-chat-latest
2. Send a few messages via WhatsApp or Telegram
3. After 2-5 turns, all responses fail with the 400 error above

Expected Behavior

Reasoning items from previous turns should either be:

• Stripped from session history before replaying to the OpenAI Responses API, OR
• Sent with their required following items intact

Actual Behavior

Stale type: "reasoning" items accumulate in the session transcript (.jsonl). On subsequent API calls, these are replayed without their required companion items, causing OpenAI to reject the request with a 400 error.

Logs

[agent/embedded] unsupported thinking level for openai/gpt-5.2-chat-latest; retrying with medium
[agent/embedded] embedded run agent end: isError=true
[web-auto-reply] [openclaw] 400 Item 'rs_...' of type 'reasoning' was provided without its required following item.

Environment

• OpenClaw version: 2026.2.15
• Model: openai/gpt-5.2-chat-latest
• Channels affected: WhatsApp, Telegram
• OS: macOS (Apple Silicon)
• Node: v25.5.0

Workaround

Manually clearing/backing up the active session file in ~/.openclaw/agents/main/sessions/ restores functionality until reasoning items accumulate again.

Related

This is a known issue with OpenAI's Responses API across multiple frameworks:

• openai/openai-agents-js#326
• vercel/ai#8031
• OpenAI Community thread