Problem
After long agent sessions or context compression, the agent’s internal reasoning (“thinking”) disappears. The chat keeps final answers and code diffs, but not why a particular approach was chosen. Why this matters
Transparency — users see what alternatives were considered and why some were rejected.
Fewer repeat questions — no need to ask again “why did you do it this way?” after a long thread.
Faster course correction — developers understand the agent’s logic earlier and can steer sooner.
Continuity on multi-step work — for architecture → implementation → tests, thinking acts as a decision log that is currently lost.
Trust — visible reasoning reduces “black box” feeling, especially when many files change at once. Proposed solution
Optionally persist thinking in chat history (e.g. a collapsible “Reasoning” section).
Settings: always show / on demand / hidden.
On summary or session resume, keep a short decision log from thinking, not only the final diff.
Not everyone needs this all the time, but the option to keep and review thinking would meaningfully improve user–AI collaboration on complex engineering tasks. Environment (optional): Cursor [your version], Linux
Hey, thanks for the detailed feature request. It’s well written, and the split between “reasoning visibility” and “preserving a decision log during compression” is helpful.
The part about controlling the visibility of thinking blocks always / on-demand / hidden, expanded by default is a direction that’s already on our radar, and we’re discussing some aspects internally. Right now, thinking is saved per message, but it’s collapsed by default, and when we do summarization or compression, earlier reasoning isn’t carried over. Your “decision log” idea that survives context compression is a separate and less common request, and I’ve noted it separately.
Definitly, this is anoying when trying to read and it collapses instantly with a fast model like composer-2.5.
Another solution could be a pause function that stops the Agentprocess to give the time to read the history (including thinking) and stop it if neccesary.