Claude Code "Extended Thinking" display is a summary, not raw reasoning

Category: Agentic coding
Status: discussion
Sources: patrickmccanna.net, discussion
Summary: Patrick McCanna wrote on 2026-06-22 that the text shown by Claude Code's ctrl+o "Extended Thinking" view is a summary of the model's reasoning rather than the reasoning itself. He reports that session logs contain encrypted signature blocks instead of readable reasoning, and that Anthropic's API documentation already states the returned reasoning is a summary; he argues the wording is indirect.
Comments: HN commenters noted OpenAI summarizes reasoning the same way and that the documented behavior is unsurprising; others debated how faithfully a summary can represent an opaque generation process.
Why it matters: Practitioners using displayed thinking traces to audit agent behavior should treat them as summaries, not verbatim chains.