A bundled chart plugin now lets Claude draw charts right in the transcript. It also charts metrics from MCP tools automatically, and the Auto charts switch in /config controls that. You can now turn off fork mode with the new forkSubagentDisabled setting. Setting includeUnsupportedModels to true lists models that Claude Code would normally hide. With claude remote-control, a cloud session can now ask to use a folder on your computer through a consent card. Programs that launch claude mcp serve can set CLAUDE_CODE_MCP_SERVE_WORKSPACE_ROOT so only folders below one root can be registered.
Several features are in this build but are not switched on yet. Groundwork for a hybrid sign-in mode now sits in Claude Code's sign-in state. There is also early work on asking your organization for approval. A marker that would let MCP tool results respect stop hooks is in place but hard-wired off. A new remote switch, off by default, would add a folder prompt option and a check on malformed permission requests.
Skills that come from plugins can now be switched off with the skillOverrides setting. Organization-level overrides are checked too. The resume picker now shows an error instead of quitting when some sessions fail to load. Passing --help to claude remote-control now prints its help instead of starting a session. Parked MCP tasks now survive an error instead of being shut down. When Claude Code runs as an MCP server, it can no longer fork subagents.
Written by our agent from the shipped bundle, not by Anthropic.
Claude Code gains hybrid sign-in state and can fetch remote managed settings with a token supplied by the program hosting it
Settings & configNew hidden flags --include-unsupported-models and --allow-other-models let you run models this version would otherwise reject
A cloud session can ask through Remote Control to use a folder on your computer, shown as a permission card, behind the tengu_violin_colophony flag
Sessions & agentsCLAUDE_CODE_MCP_SERVE_WORKSPACE_ROOT lets a host running claude mcp serve limit folder registration to a folder below the server's own
The Agent tool now explains subagent_type "fork" and when to fork, and fork mode can be turned off with the forkSubagentDisabled setting
Sessions & agentsWant the reasoning? Read walks the 26 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.
Claude Code gains hybrid sign-in state and can fetch remote managed settings with a token supplied by the program hosting it
Unclear What the host check requires is not stated, so it is unclear when --host-auth-is-org-sign-in takes effect.
New hidden flags --include-unsupported-models and --allow-other-models let you run models this version would otherwise reject
Unclear What --allow-other-models does on its own, and how it relates to --include-unsupported-models, is not shown.
A cloud session can ask through Remote Control to use a folder on your computer, shown as a permission card, behind the tengu_violin_colophony flag
Unclear The switch that controls this card is not identified, so whether it is turned on anywhere is not clear.
CLAUDE_CODE_MCP_SERVE_WORKSPACE_ROOT lets a host running claude mcp serve limit folder registration to a folder below the server's own
The Agent tool now explains subagent_type "fork" and when to fork, and fork mode can be turned off with the forkSubagentDisabled setting
Unclear Which users the server-controlled switch turns fork mode off for is not settled.
A new idleCompaction setting, shown in /config and on by default, turns off compacting a long conversation while your session is idle
Unclear It is not stated what decides whether the Idle compaction setting appears for you.
Claude Code serving its tools turns off fork subagents, and a tool host can now approve RunMonitorCommand calls through a new permission request
Unclear It is not stated how this restriction is set, or whether a reader can set it at all.
Claude Code adds internal state for organization approval requests and a new usage-reporting event, with no visible change yet
Unclear What these additions change for users is not stated.
Plugin sync can deliver claude.ai skills as plugins, and a skill replaced by a plugin is hidden, redirected, or switched off along with it
A bundled plugin lets Claude draw charts in the transcript and charts metrics from MCP tools automatically, with an Auto charts switch in /config
Unclear Whether the server-controlled switch for this plugin is on for any account is not known.
Claude Code reads CLAUDE_CODE_DESKTOP_SKILLS_PLUGIN_DIR at startup and does not pass it on to programs it starts
Unclear It is not stated what condition must hold for the folder to be kept, or what Claude Code then does with it.
Code for a marker that would let MCP tool results respect stop hooks is in place, but it is hard-wired off and does nothing yet
Passthrough and remote tool announcements now check whether a device bridge server is present, and an 'attached' result can wait for the computer to announce
includeUnsupportedModels setting lists models Claude Code would hide#Setting includeUnsupportedModels to true makes the model list include models Claude Code would otherwise filter out as unsupported
Unclear How the setting is meant to be set, from the command line or from a program using Claude Code, is not clear.
The model you pin with CLAUDE_CODE_AUTO_MODE_MODEL for auto mode's safety check is ignored, with a warning, when server or policy says otherwise
Claude Code now watches an organisation policy state while running and reloads policy settings when it changes
Unclear What the organisation 'ask' state is, and what turns it on, is not clear.
A notice for Max and Team plans about an API credit link is built in, with its wording supplied remotely and a limit on how often it shows
Unclear What the API credit link offers or where it leads is not stated.
Remote cloud sessions woken by an outside event can now hand a record of that event to the wake step, behind a server switch
Unclear What passing the event record changes about how a woken cloud session behaves is not settled.
claude mcp serve can report on tool calls and accept rules from the client#claude mcp serve gains a ReadCalls tool that reports on running and finished tool calls, and clients can pass in deny and ask rules
Unclear Whether ReadCalls is available to every client of claude mcp serve or held behind a setting is not stated.
Claude Code bundles a Claude Test plugin, run with claude-test, with author, explorer and runner agents that drive a browser
Unclear What changed in the plugin, and whether it is available to users, is not stated.
A new remote tool lets a remote client list a folder's contents, subject to the session's permission rules and with a reason given when refused
Unclear It is not clear which remote clients can use this tool or whether something switches it on.
A new reason lets some feature quietly turn itself off when you use a third-party provider or a custom base URL, with no message shown
Unclear It is not known which feature this reason turns off.
Device hooks setup gains an extra check that can stop registration even when both account flags are on
Unclear What the extra condition that shelves device hooks checks is not stated.
Claude Code changed how it decides which plugin option values from organization policy are kept, and sets none when no values pass
Unclear Which policy plugin option values are now kept or dropped compared with before is not shown.
Claude Code now lists claude-opus-5 alongside other models in a list that one of its checks looks at
Unclear Which feature or behaviour this check controls is not stated.
What would fix it?
Smaller changes and internals, grouped as the pipeline found them. Nothing is dropped, it is only further down.
4 more of these are in What probably matters to you, above.
Claude Code tells users standalone Claude Design closes by December 14, 2026, and relays server messages when a design project has closed or moved
tengu_glittery_mitten#A new gated path decides whether mcp_meta_blockable end-of-turn requests can run stop hooks, reported as query_mcp_end_turn_stop_hooks
Unclear Which MCP end-of-turn requests fall into the kind that stop hooks can now block is not stated.
Claude Code now checks who owns and can edit Windows registry managed settings before applying them, and holds them back if the check fails
Unclear When this check runs, and on exactly which systems, is not settled.
A new send-now action, bound to ctrl+x ctrl+s and ctrl+enter, stops the running turn to deliver a waiting message, marked as supported for ccr sessions
Unclear It is not clear which clients turn on the send-now capability, since it is off by default.
The Agent tool description and the main system prompt were rewritten to guide when Claude launches subagents and parallel forks
A stopped subagent's stop is now saved with retries and checked on resume, and forked-skill agents hidden behind a plugin can no longer be resumed
In a cloud session that takes managed settings from your organization, a managed settings file may only restrict and keys that loosen policy are dropped
A built-in exploring agent can now run on the smaller Haiku model in remote sessions, not just in the situations it could before
Headless and SDK sessions no longer start in auto mode by default when the server sends no value; an environment variable still forces it
The footer at the bottom of the screen can now show a warning-coloured "wrong model?" in place of its usual indicator
Unclear What makes the "wrong model?" warning appear is not settled.
The check that applies your organization's managed settings can now stop and run again mid-session when the session's link to the organization changes
Unclear It is not clear what turns on this mid-session restart or whether it is on by default.
The /model picker can now use a prepared list of options and remembers the chosen effort, saving it only if it differs from the preset
Unclear The prepared list may start empty, so it is unclear whether this changes the picker for anyone yet.
Claude Code checks one more condition before allowing auto mode, and auto mode cannot be entered when it fails
Unclear What the new condition that blocks auto mode checks is not shown.
In a session that writes nothing to disk, incoming attachments are refused with a message explaining why
Unclear Which sessions run without writing to disk is not stated.
Upload and remote file tools now apply the permission rules the session was launched with, through one shared step
A feature list now notes that /tasks lists running workflows, cloud notices gain short hints, and an easel plugin is built in
Unclear Whether these hints and the easel plugin are switched on or shown to users outside Anthropic is not settled.
The note that can keep Claude's turn going after a failed tool call now applies to more than MCP tools, if switched on remotely
Unclear Which other sources of an end-of-turn request are now covered is not stated.
/tasks for background work and dynamic workflows#Claude Code can show "Running in background · /tasks to view", or a version mentioning running dynamic workflows
/resume added to a list of commands offered in another window#A command list that already offered fast mode now also offers /resume, described as "Resume a past session"
Unclear It is not clear which window shows this list, so it is unclear where /resume now appears.
Progress summaries for subagents can now be written from condensed notes, but only for opus or fable models when the summary setting is short
Unclear It is not known whether the short setting is on by default.
Background sessions missing a shell-set gateway now say "No LLM gateway" instead of asking you to log in, and the marker no longer leaks
MCP authentication messages now come from shared helpers and can say to start a new task or chat instead of pointing to /mcp
A background session started in a different folder no longer receives ANTHROPIC_BASE_URL, and CLAUDE_CODE_AUTO_MODE_MODEL is now passed on
In remote sessions, the send-now key and an empty Enter can now push a waiting message through immediately while Claude is working
Unclear It is not clear which remote machines support sending now, so whether this works in a given session is unknown.
MCP connections now retry once, keeping track of headers Claude Code added, and check your organization's policy both before and after connecting
A specific error during --continue or --resume now prints its message and exits, including in claude -p print mode with exit code 1
When Claude Code recovers tool calls left without results by an interruption, it now builds and stores a note about the lost calls
Unclear When the note is stored and whether Claude is shown it is not stated.
Claude Code now removes environment variables found in either of two lists when cleaning up variables set through settings, instead of just one
Unclear Which environment variables the second list contains is not stated.
After a restart, Claude can get an urgent reminder that the session restarted mid-turn and its tool calls may not have run, if switched on remotely
A long MCP task that would have failed can now be let go but still checked slowly, and picked up again when its server answers
Unclear Whether this mode is on by default is not settled.
When an artifact copies files from another artifact, the server can now leave some out, and Claude Code tells you which ones
The footer under the prompt can now carry a short hint on its first row, and startup usage data now includes MCP list sizing
A clash between a plugin's plugin.json and its marketplace entry is now recorded as an error in a list instead of failing the load
Unclear Whether a plugin with conflicting manifests still loads and works is not shown.
Claude Code now waits a short grace period for background hooks of an event to finish, then cancels the rest and drops unreadable results
Unclear Whether this changes how background hooks finish or only reorganises existing behaviour is not clear.
Claude Code skips a tool call that has been marked as abandoned, so it no longer runs after the request for it was given up
Unclear It is not stated what marks a tool call as abandoned, so it is unclear which situations, such as pressing interrupt, now skip calls.
Claude Code uses one shared check to decide whether to keep or clean up a worktree, including one you entered that already existed
Unclear It is not clear what the new check tests or what its cleanup step removes from an existing worktree.
When you interrupt Claude and a running tool does not stop, Claude Code can now mark it interrupted after a grace period
Unclear The default length of the grace period, and so whether this timer runs by default, is not settled.
If your Claude Code login or plan changes, or cannot be checked, Claude Test stops its browser helper and tells you how to restart
Switching auto memory off for the session now also clears any memory pause set by a command
The command Claude Code suggests for resuming a session can now start with an environment variable setting, such as X=1 claude --resume <id>
If an ultraplan cloud session did not start from your project as it is on your computer, Claude Code shows a notice and waits for you
A path check now rejects paths longer than 1024 characters instead of 4096, and one sandbox error for plugin folders is now reported differently
Unclear It is not clear which feature uses the shorter path limit, so it is unknown which paths it can reject.
Claude Code now holds back its automatic prompt suggestions while you are spending usage credits beyond your plan's limit
Unclear It is not clear which providers or sign-in types keep prompt suggestions while usage credits are in use.
A background session no longer needs a saved worktree path to take over a worktree at startup, but only does so in a matching worktree layout
The part of Claude Code that runs without an interactive screen now also loads checks for MCP servers denied, disabled or locked out by policy
New help text says allowed web domains are set on claude.ai, with a different place for individual and Team or Enterprise plans
When the auto mode safety check is too long and the server rejects it, Claude Code now treats some cases differently and compacts only for some
Unclear Which of these over-long cases now lead to compaction, and whether this handling is switched on, is not settled.
Claude Code rewrote how it wraps long lines that contain emoji, styled text or clickable links, so line widths come out right more often
A web address that Claude puts inside backticks now appears in your terminal as a clickable link
The prompt footer can now show a timed notification or the voice indicator on its first row, chosen from one combined hint
Unclear Whether the first-row layout is switched on for everyone, or only under some setting, is not shown.
On a publish conflict, Claude is now told no edits were applied and must merge the live page into its file before sending the same publish again
The message shown when a settings hook is blocked now says it came from the organization's or the repo's settings
Claude Code's pass that makes untrusted markdown safe now treats inline code the same way it already treated links
Claude Code's built-in rules now stop the Edit tool from changing several named files inside .claude-test/runs/ folders
The internal claude-test browser plugin now has more of its tools approved in advance, so they run without a permission prompt
Grep and file listing now check each result against permissions whenever deny rules exist for the target and an extra path check passes
One skill that ships with Claude Code now tells Claude to show you a fixed message, word for word, before doing anything else
Unclear It is not known which skill this is, what the message says, or whether it is switched on for everyone.
When /design consent or revoke fails in a cloud session, the error now sends you to claude.ai/design/settings instead of /design-login
With automatic protocol negotiation, Claude Code fills in missing required fields on MCP results instead of rejecting them
When a cloud session's container fails to start, Claude Code now puts a warning in the transcript telling you to reconnect or check the logs
The footer hints about agents and interrupting agents now show only when available and until their time to rest has run out
Stored image paths in your own messages are now shown as links in the conversation, with small layout changes to the search box and prompt footer
Unclear Whether these image links can be clicked or opened is not confirmed.
When Claude asks tool search for a tool that is already loaded, it now gets told so instead of running the search again
In remote sessions, a failed progress step is now cleared when you send a new message, and waiting messages are handled differently
A short notice now tells you how to restore a stashed prompt, and a failed image paste says no image was found or suggests scp over SSH
Unclear Which display mode or setting makes the stash notice appear is not clear.
Cloud session cleanup now removes a reftable placeholder file and points an invalid branch reference at main
--resume shows the real reason when a session will not load#When --resume with a file or session ID fails in a particular way, Claude Code now shows that error instead of a generic message
The message about where a cloud session starts now names the branch it checked and says which commits and changes stay local
Claude Code changed how its terminal display reuses earlier layout when content moves and how it clears parts of the screen
If a cloud session never starts, Claude Code now says so and asks you to resend your messages, instead of blaming a disconnect
Unclear When the new computer-folder option appears in the permission prompt, and what choosing it does, is not stated.
While moving a session to the cloud, earlier answers are kept with later ones, and the prompt is cleared when the move finishes or fails
WSL_DISTRO_NAME alone#Claude Code no longer always treats WSL_DISTRO_NAME being set as proof it runs under WSL, and can fall back to checking the system
Unclear It is not clear what condition decides whether WSL_DISTRO_NAME is still trusted.
Published verbatim by Anthropic for v2.1.302. 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 63 bullets, 13 name something an entry on this page also names, 14 name something no entry here does, and 36 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_AUTO_MODE_MODEL to choose the model the client-side auto mode classifier uses; it does not change the model for server-side auto mode
Probably background-sessions-anthropic-base-url-not-forwarded-to-oth, claude-code-auto-mode-model-added-to-a-list-that-includes-cl, new-env-var-claude-code-mcp-serve-workspace-root, auto-mode-classifier-model-resolution-with-server-override-a Idle compaction row to /config for turning idle compaction off or back on; the notices shown after an idle compaction now say how to turn it off
Probably idle-compaction-the-tengu-sunny-locket-constant-is-factored, built-in-easel-plugin-chart-tool-and-auto-charts-for-mcp-me, idle-compaction-toggle-in-config /design-sync and /design-login that standalone Claude Design (claude.ai/design) closes by December 14, 2026
Probably standalone-claude-design-shutdown-notice-december-14-2026, design-sync-authorization-success-message-adds-spacing-eleme, design-consent-and-revoke-give-cloud-session-specific-error /autocompact and /effort saving to the wrong model when a gateway names Opus 4 or 4.1 with a prefix, such as anthropic/claude-opus-4-1
No entry names this claude plugin and other subcommands sometimes exiting before a queued settings or config write finished, which could lose the change or leave a .tmp file behind
Probably plugin-sync-can-fold-converted-skills-in-as-plugins #89565CLI leaves orphaned `.claude.json.tmp.<pid>.<hash>` files when terminated during config write Not planned
bin/ folder, and reporting an error when a search that finds nothing (rg, grep, findstr, git grep) is piped into Select-Object or another formatting cmdlet
No entry names this /mcp to authorize or reconnect an MCP server where that option isn't available
Probably mcp-auth-rejection-messages-moved-to-helper-functions, mcp-error-messages-vary-by-environment-start-a-new-task-or, mcp-rejected-credential-messages-moved-into-shared-helpers classifier_category missing from OpenTelemetry tool_decision events for most auto mode classifier blocks, and new tool_decision attributes pushing out tool_use_id, tool_source and tool_parameters when OTEL_LOGRECORD_ATTRIBUTE_COUNT_LIMIT is low
Probably mcp-serve-gains-a-readcalls-tool-and-repo-rule-transport claude remote-control --help not printing its usage and exiting when not signed in
Probably remote-control-subcommand-now-answers-help /design consent and /design revoke suggesting /design-login when they fail in a cloud session, where it isn't available; the message now points to claude.ai/design/settings
Probably standalone-claude-design-shutdown-notice-december-14-2026, design-sync-authorization-success-message-adds-spacing-eleme, design-consent-and-revoke-give-cloud-session-specific-error resultType, ttlMs or cacheScope
Probably mcp-modern-protocol-connections-tolerate-results-missing-req #88128[BUG] MCP tools/list and resources/list rejected as invalid when optional ttlMs/cacheScope cache hints are omitted (protocol 2026-07-28) Open
#97319[BUG] MCP client rejects valid tools/list response due to strict validation of ttlMs/cacheScope fields (Roblox Studio MCP server) Open
#88154[BUG] Microsoft Dataverse MCP connector fails with "missing required resultType" on every call in Claude Code (works fine in claude.ai chat) Not planned
claude plugin details <name>@builtin suggesting claude plugin list, which never shows built-in plugins, when the plugin isn't found; it now points to /plugin
Probably plugin-hidden-synced-skills-hiddenbehindplugin, idle-compaction-the-tengu-sunny-locket-constant-is-factored, plugin-sync-can-fold-converted-skills-in-as-plugins .git file points at a directory that doesn't exist
Nothing to match on --plugin-dir
No entry names this claude -p and Agent SDK sessions with no permission mode to start in default mode on third-party providers, with telemetry off or on a fresh install; --permission-mode auto keeps auto
Probably claude-p-continue-resume-now-aborts-on-a-specific-error subagent_type: "fork") to be on by default in claude -p, the Agent SDK, the desktop app and VS Code, as they already are in the terminal; CLAUDE_CODE_FORK_SUBAGENT=0 turns them off, and claude mcp serve never offers them
Probably attribution-header-gains-project-tag-fork-subagent-gating-r, agent-tool-schema-text-subagents-forced-into-background-by, fork-subagent-text-added-to-the-agent-tool-prompt-fork-is-o, claude-p-continue-resume-now-aborts-on-a-specific-error, mcp-serve-gains-a-readcalls-tool-and-repo-rule-transport, mcp-serve-gains-a-one-shot-workspace-root-env-var-that-can-r, fork-subagent-enablement-gains-a-settings-kill-switch-and-a, fork-subagent-type-and-plugin-cc-plugin-easel claude -p, the Agent SDK, the desktop app and VS Code now run in the foreground by default
Probably claude-p-continue-resume-now-aborts-on-a-specific-error ← for agents footer hint to hide 10 seconds after a session opens; the count of agents waiting on you still shows
No entry names this /feedback: it now points to GitHub issues instead of an organization admin
No entry names this @browser banner saying "Browser connected" when no browser with the Claude extension can be reached
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 v2.1.301.
Claude Code, interactive mode
12 prompt changes in this release could not be quoted from the build, so no entry on this page describes them.
549 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 34 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?