Session transcripts now load in a single pass again when you resume, without pausing or retrying. Claude Code no longer stops partway to let the interface catch up. It also no longer starts the read over when the file is rewritten during loading. On very large sessions, the screen may stall briefly while the transcript loads.
Live cost and token updates during a turn are now tied to a server setting, but this is not switched on yet. It affects print mode and SDK sessions. These updates only run if the server sends a reporting interval. When no interval is set, the mid-turn updates are skipped.
Background task output links are now fixed up correctly after a session moves to another folder. Claude Code repoints those links only when the move actually worked. It also uses the right paths when it does. One kind of failed session load is no longer reported as an unreadable transcript. That error now goes through the general error handling instead.
Written by our agent from the shipped bundle, not by Anthropic.
Want the reasoning? Read walks every section of this release, each entry opening to what changed and why.
Read this release → Every row →Smaller changes and internals, grouped as the pipeline found them. Nothing is dropped, it is only further down.
Reading a session file no longer pauses for the interface or restarts after rewrites, so very large resumes may stall the screen
After a session moves to another directory, Claude Code now repoints background task output links only when the move worked, using the right paths
Mid-turn cost and token updates for print and SDK sessions now only run if a server-provided interval is set; with none, they are skipped
Claude Code stopped taking a per-file lock on session transcripts when moving them, hydrating sessions and writing subagent transcripts
Unclear It is not clear whether the removed waits were real file locks, or whether equivalent protection now happens somewhere else.
Reading a session's transcript no longer pauses to let the interface respond, so resuming a very large session may briefly freeze the screen
Claude Code now reads every subagent transcript of a session at once instead of a few at a time; a failed read still just skips that one
Compacting a session transcript and deleting messages from it no longer start by waiting on the file, which may have been a lock
Unclear It is not clear whether the removed step actually locked the file or did nothing at all.
Claude Code no longer passes a fixed limit on how many subagent transcripts it loads at the same time
A session load that fails with one particular error no longer goes through the 'transcript unreadable' path; the general error handling takes it instead
Published verbatim by Anthropic for v2.1.291. Text is unmodified from the upstream changelog. Everything else on this page came out of the bundle instead, which is why the two lists don't match.
Of these 2 bullets, 0 name something an entry on this page also names, 0 name something no entry here does, and 2 name nothing specific enough to line up either way. The pairings are made on names both sides wrote down, a flag or a setting or a slash command, so read one as probably the same thing rather than as a fact, and read the middle number as candidates rather than as a miss count.
No change to the system prompt since v2.1.290.
Claude Code, interactive mode
384 documentation changes were recorded within 24 hours either side of this release, nearest first. The closest 12 are below. They're here because they happened near this release in time. That's not a claim that this release caused the edit, or that the page documents anything in it.
What's wrong with this entry?
