What does OpenClaw v2026.6.5-beta fix?

It fixes QQBot leaking raw model thinking/reasoning content into channel replies and MCP tool result coercion issues that caused Anthropic 400 errors and poisoned session history.

What is the QQBot thinking leak bug?

In earlier versions, QQBot did not strip model reasoning scaffolding before native delivery, causing raw content to appear in user-facing channel replies instead of clean responses.

What causes Anthropic 400 errors with MCP tools?

When MCP tool results return rich content types (resource_link, resource, audio, malformed images), the materialize boundary did not coerce them to text/image blocks. Anthropic's API rejected these as invalid, returning 400 errors and corrupting session history.

No. As of June 7, 2026, v2026.6.5-beta.2 is a pre-release. The latest stable version on npm is v2026.6.1.

Should I update to v2026.6.5-beta?

Only if you are experiencing the QQBot thinking leak or MCP Anthropic 400 errors. For production deployments, wait for the stable release.

← Back to dashboard

clawsmith.com/signal/openclaw-v2026-6-5-beta-qqbot-thinking-leak-mcp-coercion

📈 TrendsWide OpenLive

OpenClaw v2026.6.5-beta: Fixes QQBot Thinking Content Leak and MCP Anthropic 400 Errors

OpenClaw 2026.6.5-beta.2 strips model reasoning/thinking scaffolding before QQBot delivery, preventing raw thinking content from leaking into channel replies. Also coerces MCP tool results at the materialize boundary, preventing Anthropic 400 errors and poisoned session history from rich MCP content.