🔧 Fix: "Item of type reasoning was provided without its required following item" Error in OpenClaw

If you've suddenly started seeing this cryptic error in OpenClaw:

You're not alone. This bug (#17019) has been affecting users running GPT 5.2 and other reasoning-enabled OpenAI models since mid-February, rendering OpenClaw completely unusable for some.

What's Happening?

The error occurs when OpenAI's reasoning models (like GPT 5.2) return a reasoning block (rs_*) in their response, but something breaks in how OpenClaw handles the transcript. The API expects reasoning blocks to be followed by specific items, and when that chain breaks—boom, 400 error.

This affects:

• All conversations (every message fails)
• Cron jobs (they fail and spam you with errors)
• Browser automation and other tools

Users report it started happening spontaneously—not tied to an OpenClaw update—suggesting it may be related to upstream OpenAI API behavior changes.

Immediate Workarounds

While waiting for a permanent fix, here are your options:

1. Switch to a non-reasoning model

The fastest fix is temporarily switching away from GPT 5.2 to a model without extended reasoning:

or

Users confirm this clears the error immediately.

2. Use Anthropic models instead

Several users report Anthropic models (Claude) work fine:

The Fix Is In v2026.2.17

Good news: OpenClaw v2026.2.17 (released February 18th) includes this fix:

OpenAI/Transcripts: always drop orphaned reasoning blocks from transcript repair. (#18632)

This change ensures that when reasoning blocks are orphaned (missing their required following items), OpenClaw gracefully handles them instead of propagating the 400 error.

To update:

openclaw update

or if you're on the beta channel, the fix should already be available.

Why This Matters

This bug highlights an interesting challenge with reasoning models: they introduce new response structures that downstream tools need to handle carefully. OpenAI's reasoning blocks are relatively new, and edge cases like orphaned blocks can slip through.

If you're building tools that consume reasoning model outputs, take note—you may need similar defensive handling in your own code.

Still Having Issues?

If you're still seeing this error after updating:

1. Run openclaw doctor to check for config issues
2. Clear your session with /new
3. Check the GitHub issue for the latest updates
4. Report your findings in the thread—the maintainers are actively monitoring

This bug is a good reminder: when something breaks across all your conversations and cron jobs simultaneously, check GitHub issues—you're probably not the only one.