A critical bug affecting Claude Code v2.1.154 and v2.1.155 has been resolved with the release of v2.1.156. The issue involved the mutation of 'thinking blocks' during multi-turn sessions when using the Opus 4.8 model with extended thinking enabled. This mutation caused silent payload corruption, leading to HTTP 400 errors from the Anthropic API and breaking agentic workflows. The hotfix ensures that signed thinking blocks are replayed byte-for-byte without modification, preventing these errors. AI
IMPACT Ensures stability for users of Claude Code with Opus 4.8, preventing workflow disruptions.
RANK_REASON This is a hotfix for a specific software tool, not a new model release or major product update.
Read on dev.to — Claude Code tag →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →