You can now run gh api REST calls in sessions that do not have the GitHub CLI installed. Claude Code supplies its own limited gh and sends those calls through the session's GitHub proxy. A new command, claude daemon remote-control stop-shell, stops a background shell command running for a cloud session. Programs driving Claude Code can now read a background task's output with a get_task_output request. Compression of requests sent to the model is available through CLAUDE_CODE_GZIP_REQUEST_BODY_BLOCKS and is off by default. Admins can now set CLAUDE_CODE_TRANSCRIPT_LOCAL_GC through managed policy to turn on local transcript clean-up.
This build holds 7 things that are not switched on yet. Claude Code can offer to move you to a newer model, as a one-line notice or a dialog. A reworked way of saving your shell's functions, aliases and options for Bash commands is also in the code. Both stay off unless Anthropic turns them on remotely. Groundwork has landed for running the SessionStart hooks of plugins that load mid-session. None of this changes what you see today.
Remote sessions no longer forget their background tasks when they reconnect. A stalled live-update connection now times out and reconnects instead of hanging forever. Relaunching after /update now resumes your conversation even when Claude's last reply could not be saved. Messages from Amazon Bedrock guardrails now show up in the reply instead of going missing. A shell command that only names a web address is no longer escalated as if it touched a protected file path. /plugin-types is no longer listed among the command categories.
Written by our agent from the shipped bundle, not by Anthropic.
The /plugin help screen can show /plugin directory [query] for searching the Anthropic Directory in Discover
Claude Code now reads four new startup environment variables and passes them through to cloud sessions; they do nothing unless set
Sessions & agentsA new environment variable, CLAUDE_CODE_MCP_SERVE_TOOL_OUTPUT, appears alongside the other MCP server-mode variables
Claude Code can compress the requests it sends to the model in gzip blocks, controlled by CLAUDE_CODE_GZIP_REQUEST_BODY_BLOCKS and off by default
plugin install --replace explains that it needs a full plugin idRunning plugin install --replace without a name@marketplace id now says that id is needed and that nothing was installed
gh api stand-in for sessions without the GitHub CLITools & outputWant the reasoning? Read walks the 64 entries that probably matter to you, each one opening to what changed and why.
Read this release → Every row →Claude Code now ignores only certain expected errors while listing hooks and lets other errors surface
Unclear It is not clear exactly what this listing covers or which errors still count as expected.
Claude Code now spots the session's working folder even when it is written with different case, an alias, or a Windows network path
Loading a skill now records the path of its file, and the code that skips duplicate skills has a variable mix-up fixed
Unclear It is not clear what part of Claude Code uses the recorded file paths.
The message for an attachment over the size limit now uses the copy limit, and app links no longer pick their scheme themselves
Unclear It is not clear which link type is now passed in, or whether links you see are different.
If a plugin marketplace's catalog turns out to be a folder instead of a file, Claude Code now handles that as a specific case
Unclear What Claude Code does after detecting the folder is not described.
6 more of these are in What probably matters to you, on page 1.
A configuration Claude Code reads now has a thinking_strip field, alongside its existing role_line_scan and extract_transport fields
Unclear What thinking_strip does, and what values it takes, is not known.
Claude Code has a new feature switch that is controlled from Anthropic's servers and falls back to off
Unclear It is not clear which feature this switch controls.
Claude Code checks a new remote switch that is off unless turned on remotely, and what it controls is not visible
Unclear It is not clear what this switch controls.
7 more of these are in What probably matters to you, on page 1.
The safety check on file-path writes can now return 'settings' or 'uncertain', keeps hard denies, and adds an internalHardDeny class auto-approval can't override
Unclear It is not clear which kinds of path give the uncertain result.
Claude Code's low-level file layer gains symbolic link support through Windows reparse points and macOS calls, plus new matching error handling
Unclear It is not clear where, if anywhere, this new link handling is used yet.
Claude Code now reads a pinned version from remote configuration and uses it only when it is newer than the version you are running
Unclear What Claude Code does with that version number, for example whether it limits or triggers updates, is not clear.
A file-access layer inside Claude Code can now create symbolic links and read where one points
Unclear Which feature of Claude Code uses this layer is not known.
Claude Code now treats a session as remote Cowork when its surface is remote_cowork, and an always-yes shortcut in that check is gone
Unclear It is not clear which feature depends on this check or what the removed shortcut was testing.
Claude Code now skips two direct calls to claude.ai when the current session is not allowed to make them
Unclear It is not clear which sessions count as not allowed to make these calls.
Claude Code's check for the desktop app now reports the link scheme, and treats a claude-dev app as ready without a version check
Unclear It is not clear which Claude Code feature uses this detection.
In one mode, the coordinator's instructions can now say every message goes to the user and ask it to briefly report what it launched
Unclear It is not clear what condition picks this wording, or whether the new text reaches the final instructions.
When a hook rewrites a tool call after the server reviewed it, the local auto-mode check now judges that call, and permission checks track phase and round
Unclear It is not clear what the extra step does or how prompts change as a result.
Claude Code's secret scrubber can now hold some rules back until asked and run a quick prefilter check before each rule
Unclear It is not clear which rules are marked redact-only, so it is not clear which secrets are affected.
gh is a limited built-in stand-in#If no GitHub CLI is installed, Claude is now told that gh is a built-in stand-in that only supports gh api
Unclear It is not clear what switches the stand-in on or which sessions get it.
Claude Code has a new way to reference pasted text that skips the usual minimum-length check and marks the reference as name-only
Unclear It is unclear what uses this new reference path, so its effect on what you see is unknown.
Claude Code can now blank the contents of attached documents and remove documents whose contents are empty
Unclear It is not clear when Claude Code blanks or removes these documents.
When Claude Code refuses to send a message to another session, it now sometimes skips logging the refusal
Unclear It is not known in which cases the log entry is skipped.
A message starting with / is not sent to the model when slash commands are set to be skipped, or when it comes over a bridge and fails a check
Unclear This may only reorganise existing behaviour, and it is not clear what the further check for bridge messages tests.
A stopped background task can now end with its own status and summary instead of always showing killed and its description
Unclear It is not clear which parts of Claude Code supply a different final status.
The selectable list used in Claude Code's menus can now use a keyboard-shortcut group other than the usual one, and can be forced to scroll to an end
Unclear It is not clear which part of Claude Code uses a list with a different set of shortcuts.
Claude Code now records how long its shell snapshot takes and how it was captured, and adds an extra PATH step before Bash commands
Unclear What the extra PATH addition puts on the path is not stated.
The timing breakdown before a turn's first model request splits resume and input handling into finer phases and records why a turn resumed
The usage data reported for the signed-cache shadow check now includes how long verification took
Claude Code now records each time a startup announcement is shown, and one announcement is no longer limited in how often it appears
When Claude Code compacts a conversation, the usage data it records now includes extra fields about a refused effort level
Unclear Which exact fields are recorded is not known.
Device-hook registration telemetry now reports a separate count for each reason a hook is held back
Usage data sent when a Bash command runs or fails now includes a builtin_capture field
Unclear What condition the builtin_capture field records is not shown.
Claude Code now keeps a list of the environment variables that affect telemetry and feature flags
Unclear What reads this list and what it is used for is not shown.
When a tool from a claudeai-proxy MCP server fails, Claude Code's error report now includes that server's ID
Usage data about MCP servers now includes the server's type and id, and plugin uninstalls note when an alias was used
Unclear It is not clear which MCP event gains the new fields.
Two of the usage events Claude Code reports now carry an extra tag field
Unclear What the tag contains is not known.
The usage data Claude Code sends about web search now says whether your configuration has a web search mode written in it
Claude Code's usage data about context files now counts held files by the reason they were held
Usage data about waiting for attached files no longer says whether read-ahead was enabled or used
Unclear It is not clear whether read-ahead is now always on or has been removed.
One usage event Claude Code sends now records whether a path was a folder, when that is known
Unclear It is not clear which event carries this field.
A timing check used while Claude Code is idle or waiting on you now first consults a server-controlled switch
Unclear It is not clear what this timing check drives or what a user would see when it is turned off.
An automatic action run while Claude Code works through queued messages now depends on fewer conditions, so it can fire more often
Unclear It is not clear what the action does, or whether the removed conditions are now checked somewhere else.
/remote setup to session git rules#On self-hosted runners, session token updates now use the whole /remote setup and can add fetch rules for direct upstream repositories
Unclear It is not clear what the fetch lines do in a session.
Claude Code's check for keeping the prompt cache warm now also treats the cache as lapsed when a further value goes past a limit
Unclear What the added value measures and what the limit is were not stated.
The self-hosted runner now reads its startup command phases once and hands them to the first session it starts
Unclear It is not clear what the startup command phases do, or how a user would set them.
Claude Code's retry logic now follows a list of waits, so the number of retries and the gap between them can differ by task
Unclear The default wait list and which parts of Claude Code use their own are not known.
The small program Claude Code starts to stop your computer sleeping now gets an environment built for each operating system
Unclear It is not clear which variables the helper now receives or what it received before.
Several messages Claude Code sends to the model about blocked tools, artifact copying and permissions have been reworded
Unclear It is not clear whether copying SVG content from another artifact is now allowed, or whether SVG is only no longer named in the message.
In some cases Claude Code now skips checking whether it can reach a content host and marks that host as not having answered
Unclear What condition causes the test to be skipped is not shown.
On Linux, a sandboxed command that fails to start is now reported as spawn_failed instead of wrap_failed
Unclear What the text added in front of each sandboxed command contains is not known.
Claude Code now keeps a plugin-directory-bindings.json file recording which repository each plugin directory entry is linked to
Unclear It is not clear what decides whether the directory marketplace check passes.
Claude Code's list of community plugin marketplaces now takes in a separate list where it used to name healthcare directly
Unclear It is not clear what the separate list holds, or whether healthcare is still in it.
Claude Code's code for reading installed plugins, checking a local-only option and finding a hook's project folder was reworked
Unclear It is not clear what uses these pieces or what they change for users.
The limit on how many options a plugin's $.ui.ask question can offer is now checked by a shared function
Unclear It is not clear whether the shared function enforces the same option limit as before.
Claude Code adds a startup timing step for interactive input and new counters for plugin worker starts and crashes
Unclear It is not clear how these pieces fit together or whether they are in use yet.
When Claude Code finds a plugin in a marketplace, it now runs the result through an extra step that can replace it
Unclear What the extra step does to a found plugin, and when it returns a replacement, is not known.
The step that installs a plugin's npm or bun packages takes an extra input and checks lockfiles through a shared check
Unclear It is not clear what the extra input changes about how plugin dependencies are installed.
The automatic plugin update pass now keeps counts of directory updates that succeed, fail, get stuck or change identity
Unclear It is not clear whether these counts are actually sent anywhere.
When fetching part of a plugin's git repository, Claude Code now always passes --no-checkout and adds --filter=blob:none only for some URLs
Unclear It is not certain whether git really missed these options before, or which URLs get the filter.
Plugin auto-update results now count directory updates that succeeded, failed, got stuck or changed identity, and hook refusals record their kind
Unclear It is not clear whether these counts are shown to users or used only as telemetry.
Remote session lists now include a status for the machine running each session's tools, and a first accepted remote tool call is signalled
Unclear It is not clear whether anything on screen uses this status yet.
Claude Code now records whether incoming hook and plugin messages pass a sender check, but still lets them all through
Unclear Which sessions actually meet the forwarding conditions, and so get counted, is not known.
When a turn is handed off to a remote session, Claude Code can now pass along a way to place files by name, behind an internal check
Unclear It is not clear which feature check controls this or when it passes.
A background session now remembers when its connection is rejected for an authentication or reply problem
Unclear It is not clear what Claude Code does with the recorded rejection.
A background session can now check its live status to say whether it is waiting, and counts as waiting if the status can't be read
Unclear It is not clear what uses this check.
In remote sessions, the transcript row that marks a compaction is now held back and uploaded together with the summary
Unclear It is not clear whether this holding back always applies or only under some setting.
Remote sessions now answer unknown or malformed control requests with an error, and ignore stale or repeatedly replayed ones
In remote sessions, some transcript entries now wait for a compaction summary before upload, and those still waiting at close are dropped or counted
Unclear Which entries get held back is not known.
Remote sessions without a screen now tell their stall check when a permission prompt opens and when it closes
Unclear It is not clear whether anything in this build reports permission prompts to the stall check yet.
When Claude Code saves Claude's replies to the session transcript, it now adds extra details to each entry
Unclear It is not clear which details are added to each saved reply.
Hooks from cloud sessions that run on your device now trigger a periodic background refresh of your account settings
Unclear How often the refresh runs is not known.
Modules that provide hooks can now retry ones that failed to load, and the hook file watcher can be given more paths
Unclear It is not clear what, if anything, a user sees change from this.
Whether the hook modules from installed plugins go first is now decided only by managed settings, a Team or Enterprise organization, or prepended plugins
Unclear It is not clear whether the removed off switch now applies somewhere else.
Some warnings from hooks run in the cloud now have extra text added, and hook usage data records one more yes-or-no field
Unclear What the added warning text says is not known.
Type definitions for plugin function hooks now receive the list of tools, and setup failures now log the error code
Unclear It is not clear whether hook authors see any difference beyond the extra error code in failure messages.
When a plugin hook worker crashes, Claude Code's report now names a cause and a likely culprit
Claude Code now counts how many times it starts the background worker that runs plugin hooks and records that count
Usage records for loading plugin hooks modules now name the plugin, and unloaded plugins are left out of reload checks
Unclear What counts as an unloaded plugin here was not stated.
Claude Code checks a new remotely controlled switch and reads its settings from a renamed one, though what either controls is unknown
Unclear What either switch controls is not known.
A record describing the environment now accepts entries without the shell and OS version fields
Unclear What this environment record is used for is not known.
Claude Code adds internal timing steps and keeps count of answers from the browser that arrive while disconnected
Unclear It is not clear whether either change affects anything you see.
New usage events, a plain message when a file name already exists, a changed model ranking and a connection timeout option
Unclear It is not clear which of these changes a user can see.
Claude Code now clears one more environment variable for the programs a session starts, and session state tracks a tools machine
Unclear It is not clear which environment variable is cleared or what the tools machine is used for.
Looking up a conversation thread now checks whether the network is switched off first, among small Remote Control changes
Unclear What a reader would notice from any of these changes is not shown.
A value inside Claude Code that used to be empty now comes back as 10,000
Unclear It is not known what this value controls or where it is used.
Claude Code now tracks rows trimmed from the top of the transcript display and remembers steps that failed to load
Unclear It is not clear whether either change has any effect a user would notice.
Claude Code now drops any beta feature found on a separate exclusion list from the betas it requests for a model
Unclear What fills the exclusion list is not known.
When the API refuses the header for the inline-tools or late tool-additions beta, Claude Code now records an extra piece of state
Unclear It is not clear what the extra state holds or how long it lasts.
An Authorization header set in ANTHROPIC_CUSTOM_HEADERS is now read by a shared parser, and a self_hosted remote environment type is recognised
Unclear What Claude Code does differently when the remote environment type is self_hosted is not shown.
Claude Code now spots a rejected beta header from the API's own error wording rather than from the header name alone
Unclear It is not clear what Claude Code does differently once it detects a rejection.
Claude Code now reads ANTHROPIC_CUSTOM_HEADERS through a single shared function instead of separate inline code
Unclear It is not clear whether the shared function reads any value differently from the old line-and-colon split.
Claude Code's check for a rejected beta header on a 400 error now ignores two more kinds of error
Unclear Which two kinds of error are now skipped is not shown.
The settings for memory extraction now include a thinking_strip value that falls back to "window" when invalid
Unclear It is not clear what the possible values do or which memory feature these settings belong to.
--environment error message reworked#The code that loads nested CLAUDE.md files was simplified, and the error for combining --environment with --cloud is now built in parts
Unclear It is not clear whether the set of CLAUDE.md files that actually load has changed.
Memory files that were announced or loaded from subfolders now count as already in context when Claude Code checks for memory updates
Unclear It is unclear whether the /plugin-types command was removed or only moved.
The check that stops a folder's CLAUDE.md from being added twice now compares contents and reads the file's timestamp first
Unclear It is not clear what turns on the mode that shows the full path.
The prompt that picks which saved memories are relevant now asks for a reply made of a single JSON object listing the chosen files
Unclear It is not clear whether a formal output format is also enforced alongside the written instruction.
Claude Code now records when a remote tool answer is too old and when no slots are free for a remote request
Unclear It is unclear whether these are only new log labels or also change what happens to the requests.
Claude Code's routine for profiling the machine it runs on can now be asked to look for extra tools and to cap how many it lists
Unclear What will ask for the extra tools is not shown.
The device bridge now waits for its tool groups to finish reading before start-up completes, and can gather a list of what is on the machine
Unclear What decides whether the machine inventory is gathered is not stated.
Published verbatim by Anthropic for v2.1.287. 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 106 bullets, 14 name something an entry on this page also names, 40 name something no entry here does, and 52 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.
/plugin enable cc-plugin-you-should-know@builtin (for first-party sessions with telemetry on)
Probably plugin-types-removed-from-the-command-category-map, cloud-session-sync-and-memory-update-tweaks, plugin-background-update-failure-now-reported-for-anthropic, disabled-marketplace-notice, plugin-cloud-session-hooks-consent-display-and-project-sett, plugin-type-contract-now-laid-for-dependent-mods, plugin-types-generated-header-comments-reworded-mcp-types, plugin-screen-telemetry-and-marketplace-network-location-mes, plugin-not-found-errors-now-explain-removal-from-the-marke, plugin-marketplace-dedup-across-directories, plugin-types-removed-from-the-open-state-map-skill-folder, plugin-built-in-marketplace-can-be-turned-on-or-off, plugin-help-lists-a-new-plugin-directory-command, public-plugin-directory-search-and-marketplace-catalogue, plugin-install-replace-now-has-an-error-message-requiring, plugin-tip-advertise-enabling-the-you-should-know-side-ag, switch-installed-plugin-copy-confirmation-in-plugin, plugin-marketplace-community-managed-and-pinned-commit-deta n:<text> filter to the agents view that matches session names and tasks; a filter now shows matches in collapsed sections and Enter opens the first match
No entry names this prompt_text to the OpenTelemetry user_prompt event, a copy of prompt for backends that nest dotted keys; drop or mask it wherever you drop or mask prompt (anthropics/claude-code#70763)
Probably user-prompt-telemetry-now-built-by-a-helper-that-spreads-fie gh api (REST only) for sessions that use Anthropic-managed git on macOS and Linux machines where the GitHub CLI is not installed
Probably github-gh-stand-in-described-in-session-proxy-prompt, built-in-gh-api-stand-in-for-sessions-with-no-github-cli asyncRewake waking Claude over and over with "found issues" notifications when the hook's script file is missing; the broken hook is now reported once
Probably asyncrewake-hook-with-an-unopenable-script-reports-once #96087[BUG] Windows desktop (MSIX): plugin hook scripts invisible to externally-spawned interpreter — every hook ENOENTs and asyncRewake Stop hooks loop forever Open
#95727Desktop app: dangling plugin hook (security_reminder_hook.py missing) fires endless 'Background security review found issues' notifications Open
#96148[BUG] Failing asyncRewake Stop hook causes unbounded model re-wake loop (high token burn) Open
#94634[BUG] Security-guidance plugin hook file missing from session snapshot, causes repeated "No such file or directory" notifications Open
#92234[BUG] Broken security_reminder_hook.py plugin persists after reinstall Open
availableModels list, which could collapse /model to one Opus row
Probably managed-settings-denial-message-for-model-checks /model on a claude.ai login saving the current version's id, so your saved default now follows the newest Fable like Opus and Sonnet do
No entry names this /model, opusplan) rewriting earlier MCP tool announcements, which could drop earlier extended thinking
No entry names this rm (such as one on / or the home directory) losing its always-ask safeguard when the same command also redirected output to a ~ or wildcard path
Nothing to match on claude -p and SDK sessions repeating a model fallback on every later message after the model was switched while a reply was running
No entry names this claude agents after the agent exited and removed the worktree the session was started in
No entry names this /advisor pairing checks: Sonnet 5.5 can now advise Opus 4.7 and 4.8, and advisors the API would refuse are flagged up front instead of being silently dropped
No entry names this __proto__
Probably org-mcp-permission-maps-now-validated-and-null-prototype claude --teleport progress screen, and an MCP form field while it is being checked, to the screen reader again on every spinner frame
No entry names this CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS not removing the structured-output format from session-title and prompt-hook requests, which Bedrock-backed gateways reject
No entry names this --include-partial-messages sending a cut-short reply's message_stop late or never, so apps could show the reply as still in progress
No entry names this claude agents sometimes not showing the permission prompt a background session is waiting on
No entry names this /ultrareview giving advice about .git/info/attributes when the upload stops on a committed .gitattributes it cannot read, such as one saved as UTF-16
No entry names this claude remote-control failing to register behind an HTTP proxy with a misleading "Check your organization permissions" error (anthropics/claude-code#97352)
No entry names this #97352Remote Control cannot register behind an HTTP proxy: some requests are sent as absolute-URI instead of through CONNECT Closed
claude agents changing every second in screen reader mode; they now change at most every 10 seconds
No entry names this API Error: 401 instead of "OAuth token revoked"; in -p mode the error now starts with "Failed to authenticate"
No entry names this /ultrareview upload refusals advising you to copy a variable named by a repository's settings file into your own user settings
No entry names this --output-format stream-json and the SDK not streaming the turns of a context: fork skill run by typing /<skill> as the prompt, as they do for the Skill tool's fork
Probably strict-relay-post-added-and-skill-mention-attachment-type claude plugin marketplace add --sparse and git-subdir plugin installs failing with "transport 'http' not allowed" when the repository is served over plain http
Probably built-in-plugin-directory-marketplace-add-remove-list-and #96134[BUG] Regression since 2.1.174: plugin installation from HTTP marketplace fails with "fatal: transport 'http' not allowed" Closed
--plugin-url download corrupting the session's cached copy of the plugin archive
No entry names this /desktop quoting partial output when opening Claude Desktop timed out or printed too much output; the error now names the cause
No entry names this claude remote-control stopping mid-turn when the Mac went to idle sleep
No entry names this claude hanging or crashing with "Raw mode is not supported" when its input is piped or redirected; it now says why and exits (use -p for piped input)
No entry names this CLAUDE_CODE_SKIP_*_AUTH sending a different Authorization header than real requests when ANTHROPIC_CUSTOM_HEADERS repeats it
Probably custom-headers-parsing-moved-to-shared-helper, self-hosted-remote-environment-and-custom-header-auth-parsin /config: settings that cycle show ‹ › and step both ways with ←/→, narrow terminals stack each value under its label, and PgUp/PgDn page the list
No entry names this /memory: the left and right arrow keys now flip its on/off settings, such as Auto-memory
No entry names this /skill names typed mid-message: Claude is now told they are skills, including disable-model-invocation ones
No entry names this ~ target too
Nothing to match on [1m] suffix (CLAUDE_CODE_DISABLE_1M_CONTEXT=1 keeps 200K)
No entry names this claude agents to arrive as queued messages; slash commands other than /stop sent while a turn is running now run when it ends
Probably remote-tool-calls-background-command-ownership-and-fleet-s Bash allow rules and allowing hooks to prompt for, not run, shell writes to files Claude Code's file tools refuse outright (the Anthropic profile store, the host credentials file)
No entry names this alwaysLoad: false to defer all of that server's tools behind tool search
Probably tengu-mcp-legacy-url-elicitation-default-flipped-to-true-mc CLAUDE_AX_PREPARK_MS=50 to restore the pause
No entry names this /usage or /context command opening the extension's dialog instead of running when picked from the command menu
Probably context-in-hosted-workers-omits-message-breakdown A model matched these bullets to the GitHub issues they fix, so a link can be wrong.
No change to the system prompt since v2.1.286.
Claude Code, interactive mode
1187 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 57 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?