Claude Code's "extended thinking" is a summary- not authentic thinking
Claude Code records each session to disk. Those logs include “thinking blocks” — the model’s own reasoning as it works.
I went to inspect that reasoning this weekend and found a signature (600 characters long) and no text.
So I read the docs: https://platform.claude.com/docs/en/build-with-claude/extended-thinking
Matt Green looked into this and has some more detailed observations on the signature blocks.
This is worth knowing before you promise anyone an audit trail. Also- BEWARE: T he “extended-thinking” output from ctrl+o is a summary of Fable/Opus’ thinking. It isn’t the actual thinking that drove the model’s actions in a session- but a summary of the thinking logic. This is like using saving a jpeg as a .bmp and then editing the .bmp and presenting it as a .jpeg. The conversion produces data loss.
I’m underwhelmed by how Anthropic is presenting the behavior of their application. If you ever need a record of the logic a used by YOUR AGENT during a session: