You can now set CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS to make a request fail instead of waiting past that time after hitting a usage limit. This release has 272 entries, and 150 of them are usable today. On 1M-context models, a new dialog can offer to auto-compact long conversations sooner. If you accept, the lower size is saved to your user settings. The setting for showing artifacts beside tool results is now CLAUDE_CODE_DAPPER_CUPCAKE, and the old CLAUDE_CODE_SQUISHY_NEWT is ignored. Programs built on the SDK can now block up to 64 MCP tools for a single turn, and they can load more plugins into a running session without a restart.
There are 12 entries in the build that are not switched on yet. Claude Code can build shorter instructions for the Read, Grep and Agent tools, but none are in use. Rewritten instructions for AskUserQuestion, the tool Claude uses to ask you questions, are prepared but not used. A slimmed-down prompt variant sits behind a remote switch that is off by default. Groundwork for routing notifications and for spotting interface freezes is present but changes nothing yet. The browser setup prompt in the desktop app and on claude.ai now uses a newly named switch, so any earlier rollout starts over.
Long-running MCP tasks now wait and retry when the server rejects their ID with error -32602. Before this fix they failed at once. Stopping a background command whose output is saved elsewhere now points the notice to that file. Sending a session to the background mid-turn now keeps pending task notices and a scheduled-job note. Checkpoint commits now skip with a git_error reason when Claude Code cannot tell whether a git operation is in progress. Archiving remote sessions from the session list is removed, and the remove option that replaces it is not switched on in this build.
Written by our agent from the shipped bundle, not by Anthropic.
New CLAUDE_CODE_ATTACHMENTS_IN_USER_DATA variable lets remote sessions keep attachments in a user-data folder, with retention checks and hardened writes
The 'beside tool results' behaviour now reads CLAUDE_CODE_DAPPER_CUPCAKE; the old CLAUDE_CODE_SQUISHY_NEWT variable no longer does anything
The SDK reload_plugins request takes add_plugins to load extra plugin folders without restarting, refused in cloud sessions and under disableSideloadFlags
ExtensionsPrograms hosting Claude Code can set ask or deny rules for individual tools on their in-process MCP servers, at initialize or via mcp_set_servers
ExtensionsA per-turn disallowedTools list now blocks named MCP tools for one turn, feeds deny rules, and suppresses misleading tool-revoked notices
ExtensionsWant the reasoning? Read walks the 38 entries that probably matter to you, each one opening to what changed and why.
Read this release → Every row →Anything you can use today, anything that visibly changes, and anything worth poking at. One line each, open for detail.
New CLAUDE_CODE_ATTACHMENTS_IN_USER_DATA variable lets remote sessions keep attachments in a user-data folder, with retention checks and hardened writes
The 'beside tool results' behaviour now reads CLAUDE_CODE_DAPPER_CUPCAKE; the old CLAUDE_CODE_SQUISHY_NEWT variable no longer does anything
A new 'We recommend compacting earlier' dialog can offer to lower your autocompact threshold, shown only when a server-side config selects the offer
The SDK reload_plugins request takes add_plugins to load extra plugin folders without restarting, refused in cloud sessions and under disableSideloadFlags
Programs hosting Claude Code can set ask or deny rules for individual tools on their in-process MCP servers, at initialize or via mcp_set_servers
A per-turn disallowedTools list now blocks named MCP tools for one turn, feeds deny rules, and suppresses misleading tool-revoked notices
Unclear It is not shown which interface lets a caller send this list.
Cloud sessions now go through removeRemoteSession, which reports a result summary but in this build always fails with gate_off; the list also gains paging
Unclear It is not clear whether removing fails in every app that shows the session list or only in some.
Claude Code now tells managed (organization) hooks and env apart from those a person's plugin wrote, including statusLine and fileSuggestion commands
Unclear It is not clear what the extra step does to hooks that are not your organization's, or whether you will notice any difference.
Hosts can get raw MCP results with result_not_for_model, and the original working directory is left out under HIPAA-style restrictions
The list of prompt variants Claude Code accepts now includes three options without a scratchpad
Unclear It is not clear whether a user can choose these variants or only the server can.
The session resume list can show a one-line footerNote, prepared for a 'starts in cloud' hint whose only caller leaves it off and whose text is empty
Bash can force commands into the background with a keptBackgroundShells starter, and MCP results flagged not for the model bypass output limits
Automatic memory saving gains a chat-prose gate, a review window, an idle-flush minimum and a first-look pre-screen, all set by its config
Unclear It is not clear which feature these settings drive, whether it is running, or whether the default values switch any of it on.
Set CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS to make a request fail instead of waiting past that time on a usage limit
Claude Code can now work out which app it is running in and add a section saying so to its instructions to Claude
Unclear It is not clear whether this section is sent to Claude in any session yet.
After you run a command from a mod you are building, Claude now replies in a line or two and may offer the plugin-authoring skill
ReadProjectUploadText#The tool for reading a project upload is now called ReadProjectUploadText instead of ReadProjectUpload, and three more tools join it
The resume picker now knows whether new sessions start in the cloud by default and can show a note at the bottom of its session list
Unclear What the footer note says is not known.
Claude Code now passes two new remotely controlled switches to the VS Code extension, one for math rendering and one for a restore breaker
Unclear What either switch does inside the VS Code extension is not known, particularly the restore breaker.
Text inputs in plugin-built UI accept a rows option; two or more rows give a multi-line field where Shift+Enter adds a new line
Unclear The setting is marked internal, so it is not clear whether plugin authors are meant to use it yet.
mcp serve can track and stop background shell commands#When Claude Code runs as an MCP server, the connected app can now see and stop shell commands left running in the background
Unclear It is not clear what decides whether mcp serve announces this capability.
Claude Code now reads a marketplace entry file kept inside a plugin's own folder and stops with a clear error if the file can't be used
Unclear The name the marketplace entry file must have is not given.
A slash command can now pass along a follow-up prompt that makes Claude respond after the command finishes, instead of always stopping
Unclear No command is named that uses this, so it may not affect any command yet.
A new server-set option, inFlightWindowMinutes, can extend the keep-alive window while a run is in progress; it does nothing until set
Claude Code now contains a longer alternative version of its instructions on how Claude communicates, used only when a check passes
Unclear What switches this alternative version on is not known.
The Agent and Skill tool instructions can now take extra options with no visible effect yet, and artifact publish edits are now recorded
Claude Code can now build alternate shorter instructions for the Read, Grep and Agent tools, but none are in use yet
Claude Code can now choose rewritten instructions for AskUserQuestion and for memory upkeep, but the rewritten versions were not seen in use
Claude Code's system prompt can now be built in an alternate layout with extra sections, but the prompt sent in ordinary use is unchanged
Unclear It is not clear what turns the alternate system prompt layout on.
Environment-change reminders sent to the model can now include a notice when an agent proxy first appears
Unclear The finding does not say what the agent proxy is or when it appears for a user.
The description of the environment that Claude Code gives the model can now include a different line when an agent proxy is in use
Unclear It is not clear what turns on agent proxy mode.
Claude Code adds internal tracking for a lean rewrite mode and for background commands that are kept
An unapproved session can now be given a third outcome, 'cloud', alongside the existing 'card' and 'review'
Unclear What the 'cloud' outcome checks for and what it does for the user are not stated.
CLAUDE_PLUGIN_DATA now fills in plugin agent, command and skill prompts, and fails with an error when the plugin is on another machine
Unclear The variable name a plugin author writes to refer to the data folder is not stated.
An MCP tool call made through the control channel can mark its result as not for the model, and too-large results now return an error text
Unclear It is not clear what turns the not-for-model setting on or whether any program sends it yet.
Remotely managed settings now accept keys such as effortLevel, outputStyle and attribution, and Claude Code's settings report includes the output style
% to jump to the matching bracket#In vim mode, pressing % now moves the cursor to the bracket that matches the one under it
% after an operator#In vim mode, % can now be used as the motion after an operator, the way j and k already could
What would fix it?
Smaller changes and internals, grouped as the pipeline found them. Nothing is dropped, it is only further down.
8 more of these are in What probably matters to you, above.
Servers from --mcp-config can now be kept despite the allowlist with a notice shown, and connection options are passed through to them
Unclear The finding does not say what these connection options contain or what sets them, so any effect on how servers connect is unknown.
When /model's check hits a five-hour or seven-day limit, the result is flagged sharedLimitReached and the switch can go ahead instead of failing as invalid_model
project_write now allows local_path for text files up to 1 MiB in set folders, refuses notebooks, and explains each refusal more clearly
Without a plan file, plan mode now tells Claude to give the plan as text, and plan references show 'Plan referenced' instead of a path
Diskless sessions no longer write debug log files even with CLAUDE_CODE_DEBUG_LOGS_DIR set, and the debug logger and v5 storage warnings change
Messaging or resuming a subagent now counts against a limit that resets when you write, and is refused with a 'Not sent' message once used up
Unclear It is not clear in which sessions this limit is switched on.
If your organization restricts MCP servers, Claude Code's own server in remote sessions keeps only its repository tools, without routines or PR tools
Sending a message to a running agent now fails when the session is full, and resuming a stopped agent can be refused on a conflict
Remote Control sessions now let the hosting app send internal details and read the session ID with the worker's connection number
The managed CLAUDE.md is no longer always loaded, and the user CLAUDE.md now also needs a user file path to exist
Unclear The condition that now skips the managed CLAUDE.md is not known.
When there is no user settings folder, Claude Code now skips your personal CLAUDE.md, rules, skills and memory instead of looking for them
Unclear It is not clear what makes Claude Code run without a user settings folder.
Subagents run by workflows are now told that a note saying their task has no user authority is not a reason on its own to skip it
In a session without local storage, Claude Code now refuses to write transcripts and other session files directly to disk
Autofix PR now starts its remote session without passing the pull request's owner, repository and number
Unclear It is not stated whether the remote session still shows or links to its pull request some other way.
When another program handles sign-in and only a token from an environment variable is in use, Claude Code now still fetches and re-checks policy limits
Unclear Which setups count as a host managing sign-in is not clear.
Claude's instructions for building a phone-friendly page now say how to fill the strip above a sticky header near the phone's notch
Unclear Whether these page-building instructions are in use for everyone is not clear.
In sessions with no local disk, transcript entries with nowhere to go are now dropped with an error, and metadata folders are no longer created
If the normal security check fails during an administrator's gateway login, Claude Code now checks a customer allowlist before giving up
Unclear It is not known where the allowlist comes from, or whether it holds any entries in this version.
Claude Code adds a note telling Claude to continue after its worker restarts, and an error when thinking from another organization can't be used
In a certain mode, Claude Code no longer looks in your home folder for skills, agents and permissions, and file writes always need approval
Unclear It is not clear which mode or condition causes these home folder lookups to be skipped.
When the wait is skipped, the account plugins freshness check now runs in the background and saves its result for later turns
Claude now gets guidance on explaining that a chat isn't linked to a computer, or that Claude in Chrome is off
If an MCP tool asks to end the turn while a tool call in the same batch failed, Claude Code can now keep the turn going
Unclear Which kinds of MCP tool failure count, and what the added note says to Claude, are not stated.
$.env.set now return a deny result#The handler that warns about a $.env.set call now also returns a deny result instead of returning nothing
Unclear Whether the deny result now actually blocks a $.env.set call is not stated.
If a session has no config folder to check a file path against, Claude Code now refuses the write, and refuses a retry too
When a session is too full to resume, Claude Code now stops and reports it instead of carrying on
Unclear It is not clear what makes a session count as full or what the message says.
Claude Code can now refuse to read a particular file because it changes what the child loads from that folder
Unclear It is not clear which file triggers this refusal, or whether the shorter list of reasons is part of the same change.
The plugin publish message now says the new version was saved, nothing was sent to your organization, and to publish from claude.ai
If a plugin upload is not its current version, Claude now tells you to make it current and publish it from claude.ai
The artifact tool description stops mentioning root, edits with root are no longer rejected, and MCP results note result_not_for_model
Unclear It is not known whether result_not_for_model can be used yet.
The message view now shows pending (queued) prompts after the transcript and keeps track of the latest compaction summary
Unclear It is not clear what condition makes the pending prompts or the compaction display appear.
After a context cut or rewind, Claude Code also clears file paths as they were typed and lets removed skills be listed again
After a conversation is compacted, the notice can now say where the earlier messages are saved
On exit, Claude Code checks whether a worktree moved to another branch and only removes it on its own when no commits would be lost
Unclear It is not shown whether the check for a worktree that has moved branch is switched on by default.
Permission rules in user settings that start with / are no longer resolved against a root folder in the permission check
Unclear It is not clear under which condition this applies, so whether your absolute-path rules match differently is unknown.
Models whose 1M-token window is only a beta now get the default window unless the 1M opt-in applies
Unclear Which models are affected is not stated.
When a session has no memory folder, Claude Code refuses to write any file whose content looks like it contains secrets such as passwords or keys
In sessions with no local disk, reading stored transcripts and writing teammate messages now fail clearly unless a remote store is available
Claude Code has a new error, "This session has no config or temp directory.", for sessions that have no disk folders
When Claude runs a shell command in the background, Claude Code can now point Claude at a specific output file instead of the default location
Unclear It is not stated what causes background shells to be kept track of separately, so when the new output path applies is unknown.
Background agents carried over from an earlier session that Claude Code cannot confirm are now marked failed, and Claude is told why
In a cloud session, Claude is now told that settings-file edits need approval on the machine running Claude Code, and not to retry
Problems in sandbox and sandbox.credentials settings, including warnings, are now judged by their own check for whether they stop settings from loading
Unclear It is not clear whether any settings that loaded before will now be stopped, or the other way round.
When Claude Code starts a subagent, it now adds a system message to the conversation
Unclear It is not stated what the message says or whether you will see it.
Claude Code's queue of staged input can now mark entries that are never dropped when the queue gets full
Unclear It is not stated which input gets marked as never dropped.
Messages about a branch missing from GitHub or ahead of its remote can now leave out the 'push it' advice, controlled by advisesPush
When Claude Code syncs plugins, it can now re-read plugin state during the session instead of relying on a single fetch
The agent list in the terminal can end with a line pointing to older sessions, and in some cases no longer scrolls to the focused item
Unclear Which situations stop the list from scrolling to the focused item is not known.
When automatic replies to artifact comments do not resume, Claude Code can now show a short message pointing you to /tasks
A usage line reading "<used> of <limit> used" now ends with either "Resets" or "Expires" followed by a date
Claude now suggests the Claude desktop app or Claude in Chrome when no browser is connected, and no longer gives a link
When MCP servers are filtered by policy, Claude Code now notes the ones it kept and says deniedMcpServers in managed settings removes them
On Windows, Claude Code now reads settings from the per-user registry only when an extra condition also holds
Unclear It is not clear what the new condition checks, so it is not known when registry settings are now skipped.
If your organization defines SessionEnd hooks, CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MS is read the organization's way, not straight from your setting
Claude Code now uses a different limit on how many rows of output to show in the desktop app, VS Code and mobile
Unclear Whether the new limit shows more or fewer rows than before is not stated.
On self-hosted runners, failed session releases now log how many failed in a row and always restart the idle timer
A finished background task now reports either its exit code or that it was killed, with an output-limit reason when known
Unclear It is not certain that this reporting belongs to background tasks rather than some other part of Claude Code.
The description of the mcp_call request now says results skip the usual model-side processing when result_not_for_model is set
Unclear Only the request's description mentions result_not_for_model; whether requests already accept it is not confirmed.
Claude Code can now tell when a prompt is a slash command it does not know, and counts relayed prompts separately
Claude Code now adds output styles found in other project locations to the ones it loads from your project settings
Unclear It is not clear which other locations these extra project output styles are loaded from.
Claude now gets explicit notes when a failed tool call stops a turn from ending and when an oversized MCP result is dropped
Vim mode and the up/down history keys can now be switched off while input is inactive, and vim handles Esc-then-key sequences differently
Published verbatim by Anthropic for v2.1.301. 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 51 bullets, 8 name something an entry on this page also names, 15 name something no entry here does, and 28 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.
CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS so unattended retry mode (CLAUDE_CODE_RETRY_WATCHDOG) ends a request at a usage limit that resets too far away instead of waiting
Probably new-env-var-caps-how-long-the-retry-watchdog-waits-on-a-usag claude-api skill, including setting up trace destinations with the ant CLI
Probably tip-for-multiple-sessions-is-hidden-in-some-cases-the-claud % motion to vim mode: jump to the matching bracket, alone or with d/c/y
Nothing to match on rows to a mod's Input: a text field of several lines in the terminal, where Enter submits and Shift+Enter or a backslash before Enter starts a new line
No entry names this classifier_verdict and permission_mode to OpenTelemetry tool_decision events, so auto mode classifier decisions can be told apart from permission rules
Probably auto-mode-classifier-verdict-added-to-analytics, tool-decision-telemetry-gains-permission-mode-and-config-dec classifier_category to OpenTelemetry tool_decision events, naming the built-in rule a classifier block matched, when OTEL_LOG_TOOL_DETAILS=1
Probably auto-mode-classifier-verdict-added-to-analytics, tool-decision-telemetry-gains-permission-mode-and-config-dec WorktreeCreate hook without asking when it may hold unpushed commits
No entry names this fable (and best) model alias erroring on the Anthropic API when the newest Fable model is not available to your organization: Claude Code now retries once on the previous Fable model
Nothing to match on >0q covering the folder trust prompt, and >4m before the shell prompt at exit, in terminals built on VTE older than 0.53 such as MATE Terminal on RHEL 8
No entry names this GOK text appearing in the prompt at startup in some versions of iTerm2
No entry names this #100691[Bug] Stray "GOK" appears in prompt at startup in iTerm2 only Open
#99985Text "GOK" appears in the prompt at startup in iTerm2 (kitty graphics query reply not consumed) Open
#98973Startup input box pre-populated with stray text ("GOK") when project CLAUDE.md/AutoMem files are loaded Open
#95448Kitty graphics probe: iTerm2's i=less OK reply fails the response regex and is typed into the prompt as GOK Open
application-inference-profile running your awsAuthRefresh command and delaying model-switch hooks when the provider is not Bedrock
No entry names this claude auth status or claude plugin list exited in the middle of a token refresh
No entry names this #95822Short-lived commands (`claude auth status`, `claude --bg …`) start an OAuth refresh at init and exit without saving it, leaving the profile with a spent refresh token Open
#102265[BUG] Short-lived subcommand (claude plugin validate) exits mid OAuth refresh: the rotated refresh token is never saved, the next process gets invalid_grant and blanks the login Open
/rewind
No entry names this ask decision not showing which plugin or skill the hook comes from
Nothing to match on /diff not noticing a conflicted cherry-pick or revert in reftable repositories when one of its git checks timed out or failed
No entry names this cd in the same session
Nothing to match on $.model.fork returning nothing-to-fork when called before the first response of a new Claude Code process had finished, as in claude -p runs and cloud sessions
No entry names this /compact and auto-compact briefly freezing the UI while the visible transcript is rebuilt
No entry names this deniedMcpServers to keep those tools off
Probably mcp-server-filtering-now-explains-servers-kept-despite-allow [killed] line in their output file
Probably background-shell-exit-marker-replaced-by-a-structured-call, background-shell-kill-output-file-handling-changes-new-aut .catch handler to run only when the hook throws or times out; events raised beneath the hook's own calls now skip that hook without asking its .catch
Nothing to match on $.env.set to resolve with {} once the variable is set or unset, or with { deny } saying why when the write is not made
Probably hook-env-set-denials-now-return-a-deny-result $$ and multi-line \[ … \] blocks and inline $…$ and \(…\) math
No entry names this A model matched these bullets to the GitHub issues they fix, so a link can be wrong.
No change to the system prompt since v2.1.300.
Claude Code, interactive mode
830 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.
The 36 literal strings found in the bundle, with the number of entries that name each one. Picking one searches for it. A name is here because this build's code mentions it, which is not the same as it working or being finished.
What's wrong with this entry?