Admins running the bundled gateway can now set one shared monthly spending limit for a whole access group. The new blockReadsExemptPaths setting lists folders Claude Code may still read when reads outside your working directories are blocked. Claude Code can now publish an artifact as a Claude doc. It also reports whether a lasting subscription to that doc's updates was set up. In Cowork, the loops Claude Code starts no longer receive your memory. A hidden --gateway-json option on claude auth login lets the VS Code extension handle Cloud gateway sign-in.
Several features are in the build but not switched on yet. Remote sessions can track and resend conversation lines they still owe after a restart, but only when this is switched on remotely. Files uploaded to a project can be attached as text context, behind a flag that defaults to off. Claude Code can list MCP server requests that are waiting for your input, also behind an off-by-default flag. Long Bash output previews can keep both the start and the end, with the full text saved to a file. Numbered menus can be set to wait for Enter instead of picking after a short pause.
File searches no longer hang when ripgrep fails to exit after a timeout. Stray connections can no longer knock you off a background session. Sending a message immediately no longer cuts off an automatic compaction that is in progress. Resuming a copied session no longer shows a stray cut-off tool call. When Claude Code runs as a tool server, a settings file that fails to load no longer drops its permission rules. Shell snapshots now save your aliases in a case where they used to save nothing.
Written by our agent from the shipped bundle, not by Anthropic.
A new held-message recognition feature can be turned off with CLAUDE_CODE_DISABLE_HELD_MESSAGE_RECOGNITION, which is also passed to remote sessions
CLAUDE_CODE_SORTED_OTTER can now override the remote tengu_sorted_otter switch, and one function moves from that switch to tengu_quiet_pike
claude auth login gains a hidden --gateway-json option that runs Cloud gateway sign-in as JSON lines for another program, such as the VS Code extension
Managed settings gain policy_mode: layered, which merges several matching policies instead of applying only the first one, with limits to protect deny rules
A background session's socket can now hold new connections until they send something, so a silent stray connection cannot kick off the attached viewer
Sessions & agentsWant the reasoning? Read walks the 37 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.
Artifact publishing gains a claudeDoc flag that can set up a durable subscription, with a new claude_doc_not_admitted failure reason
Unclear It is not clear what triggers Claude doc publishing or whether users can reach it in this release.
A new held-message recognition feature can be turned off with CLAUDE_CODE_DISABLE_HELD_MESSAGE_RECOGNITION, which is also passed to remote sessions
Unclear It is not clear what held-message recognition does or when the internal check lets it run.
CLAUDE_CODE_SORTED_OTTER can now override the remote tengu_sorted_otter switch, and one function moves from that switch to tengu_quiet_pike
Unclear What holding while connecting does for you, and what else this switch controls, is not stated.
Remote sessions gain a ledger of transcript rows not yet delivered to the client, resend them later, and hold urgent requests while that happens
A background session's socket can now hold new connections until they send something, so a silent stray connection cannot kick off the attached viewer
Unclear It is not settled which program's environment receives this variable, or what the remote switch falls back to when the server sends nothing.
Restricted mode now blocks reads outside the working directories, searches and Chrome uploads are fenced too, and blockReadsExemptPaths lets chosen paths through
Unclear Under some conditions the exempt list is treated as empty, and it is not clear what those conditions are.
How review findings are reported is now chosen as 'spawn' or 'message', replacing the old CLAUDE_CODE_REPORT_FINDINGS check
Unclear It is not stated what the "spawn" mode does or which tool these findings come from.
Remote session code gains renewSessionTokens and isThawing hooks, but isThawing always returns false and the function receiving them does nothing
Setting the new CLAUDE_CODE_DISABLE_ULTRATHINK variable stops Claude Code from treating the word ultrathink in your prompt specially
Unverifiable self-named and looped cross-signed CA certificates are now dropped, two env vars keep them, and certificate errors explain missing issuers
Unclear Which certificates are dropped, and whether dropping happens by default, is not settled.
smaller_pass gains a form setting#The smaller_pass block in CLAUDE_CODE_POST_TURN_MEMORY_CONFIG now takes form (turns or inline) and its mode can only be off or shadow
Unclear What choosing turns or inline changes is not stated.
Claude Code can now attach the text of documents uploaded to a project, but only where a flag that defaults to off is on
Many entries in Claude Code's list of model capabilities now include a capability named tool_loaded_text
Unclear It is not stated what this capability changes in how Claude Code works with these models.
Claude Code now carries three new built-in prompt documents and a new stamp that records the session file together with a counter
After sign-in, Claude Code can now load a dynamic MCP setup and add a server, unless strict or managed MCP settings apply
Unclear It is not clear which server gets added or what decides whether one is added.
If a Claude Doc cannot wake your session on comments, Claude Code now says @Claude comments will go to another Claude
In Cowork, Claude Code now sends empty memory to the loops it starts, and a new emergency check can turn memory off on the laptop side
In remote sessions, successfully creating a document with the artifact tool can now set a wake, when artifacts and comments are both on
Unclear It is not clear what setting a wake does after a document is created.
Claude Code's memory extraction can now run a trial step before its work and log the result after, and now reports when a direct write is skipped
Bash results can now include a note when a process was killed for running out of memory, plus hints set from the server
The part of Claude Code that runs remote sessions now tracks owed output rows and display hooks, and its old first-idle wait was removed
The description of each model option now accepts an optional early access field
Unclear It is not clear whether the model picker shows anything different for models marked as early access.
SDK apps and editor extensions now get ultrathinkAvailable, and a settings file that briefly cannot be read keeps its last known values
Waiting MCP elicitation requests are now kept in session state and sent with pending permission and dialog requests, then removed before delivery
Claude Code now passes a new variable, CLAUDE_CODE_SESSION_BEGAN_AS_COPY, on to the programs it starts
Unclear What this variable does, who sets it, and whether a user should ever set it is not stated.
CLAUDE_CODE_SUBAGENT_EFFORT sets a default effort for subagents#Set CLAUDE_CODE_SUBAGENT_EFFORT to give subagents and workflow agents an effort level when their own definition does not set one
Unclear Which effort levels the variable accepts, other than inherit, is not stated.
With CLAUDE_CODE_MCP_NO_REDIAL_AFTER_POLICY_REFUSAL set, Claude Code can stop redialing MCP servers that the sandbox network policy refused
Unclear The second condition is not identified, so setting the variable may not be enough on its own.
One of the data formats Claude Code reads now has an optional claude_doc field, ignored if it is missing or malformed
Unclear What the claude_doc field carries and whether a user ever sees anything from it is not known.
With CLAUDE_CODE_MCP_SKIP_UNCHANGED_RELIST set to true, Claude Code reuses an MCP server's earlier tool list when nothing changed
Unclear It is not clear whether another condition, besides the environment variable, also has to be met before the remembered list is reused.
The reply to a connecting program can now list MCP server questions that were sent but not yet answered or withdrawn
What would fix it?
Smaller changes and internals, grouped as the pipeline found them. Nothing is dropped, it is only further down.
5 more of these are in What probably matters to you, above.
After failed or killed commands the Bash tool can now show Claude the output's end and memory-kill notes, and analytics record programs and test results
Unclear It is not stated what decides whether the end-of-output and out-of-memory notes are switched on.
The Bash edit diff gets a new git snapshot path, a clearer note when other commands also changed the repo, and its own reminder type
The /remote-control dialog now lists current and ended shares, and its menu cannot be used while a share is active
Unclear It is not clear what is being shared or whether the feature is switched on for anyone.
With a gateway set, auto mode now shows a notice when gateway checks are not working and it is using its built-in check, which sends extra billed requests
On Linux, Claude Code now finds processes left behind by tasks when a session shuts down, asks them to stop, then force-stops any still running
In the expanded permission prompt, one approval option can now show a fixed label and refuse to be selected when a condition is not met
Unclear Which conditions disable the option, and what label replaces it, are not stated.
The sandbox now lets Claude Code write to a staging folder for MCP tool calls inside the temporary folder, and warns if it is not a plain folder
In some cases the auto mode checker now sees a trimmed conversation unless CLAUDE_CODE_AUTO_MODE_FULL_CONTEXT is set
Unclear It is not stated what the trimmed context leaves out.
When the server clears old tool uses from context, Claude Code now estimates the tokens freed and caps the reported figure at that estimate
On-device MCP serving now tracks its events in one place and sends a consent prompt again once it is no longer put off
Preview tunnels gain an option to refuse other sessions, require strict TLS and stop if the session changes, plus a report on unlisted ports
MCP servers that Claude Code's cloud service adds are now marked so apps show them as included with the session
Claude Code now only prepares a compacted conversation ahead of time while idle when the prompt cache is the one-hour kind and still warm
Auto mode now checks a remote switch, off unless switched on remotely, before approving two particular tools without asking
Unclear Which two tools are affected, and whether they are still approved automatically while the switch is off, is not settled.
In dontAsk mode a refused tool call can now say it could have been approved, and is wrapped up the same way as background-agent refusals
Prompts a plugin sends show under a 'Prompt from the plugin' header, and session titles and export filenames can use the prompt as shown
On exit, Claude Code now force-stops shell commands and process trees still running after a grace period, except on Windows
Unclear Apart from never running on Windows, what decides whether this cleanup is switched on is not clear.
The remote mode note about connected tool limits now says "managing cloud scheduled tasks" instead of "scheduling routines"
When a background task from an MCP server finishes, the notification Claude Code queues now names the MCP tool that ran it
Shell commands Claude Code moves to the background are now marked to be left alone when it exits
Unclear Whether background commands now keep running after Claude Code exits is not confirmed.
A subagent with no valid effort level of its own now gets one worked out from its permission settings instead of none
Unclear Which effort level the permission settings produce is not known.
Claude Code can now reuse an earlier tool list from an MCP server, and skips updating its state when a refreshed list has not changed
Auto mode allow rules are now gathered per settings source, and command-line allow rules are merged with managed policy
Unclear It is not stated how command-line allow rules and policy allow rules are merged.
Asking a subagent for lower or higher effort is now also turned off when a second kind of model override is set
Unclear It is not stated which setting the new model override reads.
One permission check now counts the mode that was in use before Claude Code rewrote it, such as bypassPermissions
Unclear It is not stated which feature this mode check controls.
Claude Code now uses a single size limit to decide when a large tool result is saved and replaced with a preview
Claude Code now warns when a client-hosted MCP server's permission_policy is not applied, so always_ask and always_deny have no effect
Unclear Which kinds of session show this notice is not stated.
The permission check for writing markdown (.md) memory files now also runs when writes outside allowed folders are blocked
Unclear Which markdown paths the new condition allows or blocks in the restricted mode is not stated.
The tool that lists directories from an MCP server now asks for permission or refuses, instead of only erroring when unavailable
Unclear What decides whether directory listing is switched on or off is not stated.
claude mcp serve in HTTP mode now loads remote feature settings at startup#When run as an HTTP MCP server, Claude Code now loads its remotely controlled feature settings as soon as it starts
Remote Control now refuses to start a session if the account behind its credentials changed or could not be confirmed
Claude Code now adds git ignore rules for its own files under .claude/ so they stop showing as untracked
In acceptEdits mode, a browser tool call asking only to create something if empty now runs without a permission prompt
Unclear It is not clear which browser tool this applies to.
In a hosted worker, an MCP request whose trusted CA certificates share none with Claude Code's current list is now refused with an error
When one action is waiting in a remote session, the status now reads "<tool> waiting for approval" if auto mode's check could approve it
Auto mode now keeps its "Scheduled-Task Fires" rule even when a custom allow list replaces the default rules
When a background session waits on an action auto mode's check could approve, the message now says it may be approved without you
Turn-start timing adds four new phases, and a message about artifact writes now says scheduled task run instead of routine run
Claude Code now chooses a Haiku model itself only when no small-model variable is set, the provider is Anthropic's own, and the main model is the default
Unclear It is not clear which tasks use the model this rule chooses.
Claude Code can now hold back a saved login and explain why: an admin setting, missing organization settings, or a needed restart
The description of the Bash diff setting now says it is off by default, except in auto and bypassPermissions modes under certain conditions
Unclear Only the setting's description changed, so it is not clear whether the actual default for showing Bash diffs changed too.
After a sandboxed command, Claude Code now renames planted git-repository files with a .claude-sandbox-quarantine- prefix rather than deleting them
In remote sessions, a warning now says MCP servers from --mcp-config or agent frontmatter are ignored and that --chrome has no effect
A permission reason now reads "A scheduled task's new artifact, private to the person" where it used to say routine
Shell commands moved to the background are now marked to be killed when Claude Code exits, and text output timing is now measured
Unclear Whether background commands were already stopped at exit before this release is not stated.
The timer pause for open MCP questions is now skipped for apps on a list
Unclear Which apps are on the list is not stated.
When Claude Code does not know a model, the message now reads "Requires Claude Code v<version> or later"
The waiting message now says a tool is waiting to be approved, and only says you must approve it when no automatic check can
Whether Claude Code adds its exit code 2 check to a PowerShell hook now depends on whether the script uses begin, process, end, clean or dynamicparam blocks
Unclear Whether scripts with these blocks now get the added exit code check or now skip it is not clear.
The message shown when a memory is saved can now list the written file paths in verbose mode, and its wording comes from a shared source
Unclear Whether the new path list is switched on by default is not stated.
The short label summarising an agent's work is now always written in the language of the task, for every model
The in-terminal feedback survey now lets Enter, or whatever key is bound to submit, confirm an answer
When long tool output is saved to a file, the preview Claude sees can now include the end as well as the start
When a remote session reconnects, Claude Code can now fetch only the events after the last one it has, rather than up to 1,000
Claude Code now retries an MCP connection through a proxy only when the proxy refused with a 403, and runs its test less often
Unclear It is not clear when the new check allows the test connection to run.
The Bash safety check now follows variables in relative-path checks, reads heredocs inside substitutions and flags commands naming zsh histchars
Uploads, cloud session bundles and auth checks now mark requests made with your saved login, and task uploads report egress_blocked when the network refuses
Claude Code now names three specific reasons a sign-in token fails or is revoked: account not a member, no organization, no account
Unclear It is not stated whether the message you see when login renewal fails changes.
Errors about a missing GitHub remote in ultrareview and cloud agents now share one hint, which covers a GitHub repo that already exists
Claude Code now writes its detailed log lines about setting up an MCP server's web connection only when a debug check passes
When sessionStore is used with persistSession: false, the error now advises a new private CLAUDE_CONFIG_DIR deleted afterwards, not /tmp
Downloading task files now waits for the file service connection and stops as cancelled if you abort
The notice saying a session isn't eligible for auto mode now includes a separate help link
Claude Code now flags commands where a variable set to a relative value may name a file in the working directory
The survey prompt now shows "Opening the survey…" while loading and a hint to answer by typing 1 or 0
Claude Code now recognises sign-in failures where the token has no organization, has no account, or the account is not a member
Remote sessions get several small changes, including an extra connection check and a sturdier way to spot a failed move to the cloud
Messaging an archived cloud session now tells you it can take messages again once you unarchive it at claude.ai
Anthropic has not published official notes for v2.1.315 yet. This section updates automatically when the entry appears in the upstream changelog. Everything else on this page came out of the bundle instead, which is why the two lists don't match.
4 prompts changed.
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.
629 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 44 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?