[BUG] 400 "thinking/redacted_thinking blocks cannot be modified" bricks session after large multi-tool turns
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
After an assistant turn with many content blocks (interleaved thinking + many tool_use blocks — e.g. 10+ Bash calls and/or a sub-agent spawn in one turn), the NEXT request fails with a 400 and the session becomes unrecoverable.
The malformed assistant message stays in conversation history, so every following request re-throws the same 400. The only workaround is starting a new session, which loses in-context state. This has recurred across multiple sessions running the same deploy-and-verify workflow. The offending block indices observed were content.33 and content.13 (i.e. large assistant messages).
What Should Happen?
Claude Code should resend the prior thinking / redacted_thinking blocks verbatim (or safely drop and regenerate them) so the follow-up request validates and the session continues instead of getting permanently stuck. The API's 400 looks correct — it is rejecting modified signed thinking blocks — so the fix appears to be on the client side (Claude Code's message handling on large turns).
Error Messages/Logs
API Error: 400 messages.5.content.33: thinking or redacted_thinking blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response.
API Error: 400 messages.3.content.13: thinking or redacted_thinking blocks in the latest assistant message cannot be modified.
Steps to Reproduce
- Use Claude Code with extended thinking enabled (model: Opus 4.8, 1M context).
- In a single turn, trigger a heavy batch of tool calls — e.g. 10+ Bash commands and/or spawn a sub-agent — so the assistant message contains many interleaved thinking + tool_use blocks.
- Send the next message to continue the conversation.
- The request fails with: 400 ... thinking or redacted_thinking blocks ... cannot be modified.
- Every subsequent message re-throws the same 400 — the session is stuck until a new session is started.
Note: strongly correlated with large turns. Likely cause is client-side message management (context compaction, large tool-result truncation, or block reordering) mutating the signed thinking blocks before resend.
Claude Model
Opus
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
2.1.158 (Claude Code)
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information
_No response_
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗