What changed
Claude Code 2.1.296 shipped on October 9, 2026. It adds an option for reading large files with the Read tool, a per-subagent auto-compact window, and an environment variable that sets the model for workflow agents. The fixes center on managed-settings hooks, edits to files that are not valid UTF-8, moving a session to the background with ←, and PowerShell and plugin installs on Windows. A gateway sign-in regression from 2.1.295 is also fixed. The items below are a selection of the changes with the most user impact.
New features
-
allow_largeoption on the Read toolClaude can read a text file past the usual size limits in one call when it needs the whole file and the context has room.
-
autoCompactWindowfor subagentsAvailable in subagent frontmatter and
--agentsdefinitions, so a subagent can auto-compact earlier than the main conversation’s window. -
CLAUDE_CODE_WORKFLOW_SUBAGENT_MODELRuns every workflow agent on one model while other subagents keep theirs.
-
CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MSSets a longer maximum delay for the backoff when retrying an overloaded (529) request.
-
A
codekey in the Claude apps gateway’smanaged.policies[]The same settings as
cli, also applied in Claude Desktop’s Code tab. Besidedesktop, it turns on Claude Desktop’s gateway mode. -
A note in
/pluginfor left-out hooksThe note appears on a plugin whose hooks are left out because another enabled plugin has the same name.
-
[Cloud sessions] A status filter on the Activity tab
The Sessions and Runners lists on a self-hosted environment’s Activity tab in admin settings get a status filter. You can pick several statuses at once.
-
[Claude Tag] Memory files on the Memory tab
Members with the Claude Tag Admin permission can create, edit and delete workspace and channel memory files on the Activity page’s Memory tab.
Key improvements
-
--debugoutput for hooksCommand hooks on tool calls, prompts, SessionStart and Stop now log their command, plugin, outcome and duration when they finish, so a slow hook is identifiable.
--debugalso names unrecognized frontmatter fields in custom agent files, with a hint for likely typos. -
Auto mode’s “Not run” row
A tool call that wasn’t run because auto mode’s check had no usable answer now shows as a dim “Not run” row instead of a red error.
-
Transcript toggle (ctrl+o) responsiveness
Faster in code-heavy conversations, by caching syntax-highlighted code blocks.
-
Fullscreen mode links
The link under the mouse pointer is now underlined to show it can be opened.
-
Errors for a refused cloud session
They now say why your organization’s policy couldn’t be loaded, and name an API key or auth token that overrides your claude.ai login.
-
Claude Desktop’s Code tab behind a Claude apps gateway
When a session can’t start, including on a machine not set up for the gateway’s
codesettings, the reason shows as a reply. -
CLAUDE_CODE_TRANSCRIPT_LOCAL_GCA large transcript file is no longer rewritten when that would free under 10% of it.
Changed behavior
-
Default limit on MCP tool descriptions and server instructions
The default limit on MCP tool descriptions sent up front and on MCP server instructions goes from 2,048 to 4,096 characters.
-
A turn started while
←moves the sessionA turn or
!command started while the session moves to the background is now stopped instead of finishing out of sight. -
Sonnet 5.5 cache read pricing (updated)
/cost, the status line,--max-budget-usdand the SDK’s cost figures price Sonnet 5.5 cache reads at $0.10 per million tokens (was $0.20). -
[VSCode] Claude in Chrome asks before browser actions
It now asks in every session, including one connected with
@browser, as the terminal does. Allowing a site for the session stops repeat asks. -
[Claude Tag] Footer on Slack replies
Claude in Slack shows the footer with the session link, model and cost only on its newest reply in a thread. The reply before it drops the footer.
Bug fixes
Hooks
- Managed-settings
PreToolUsehooks that deny a tool call with"continue": false, and managedprompthooks that block one, refusing the call but not ending the turn. - PostToolUse hooks in managed settings not applying
updatedMCPToolOutputin some sessions. - Esc or an interrupt during a
UserPromptSubmithook or a mod’sprompt.submithook ending headless sessions, clearing the typed prompt, or letting the unchecked prompt through. - Hook output being altered when it contained text resembling a plugin hint tag.
- PowerShell commands not seeing variables that hooks write to CLAUDE_ENV_FILE (when the file holds only plain assignments).
- SessionStart hooks of a plugin loaded after a headless session starts being skipped when a different plugin with the same name, or another spelling of it, had already run.
File edits & permissions
- Edit and NotebookEdit replacing every non-ASCII character in files that are not valid UTF-8 (Windows-1252, Shift-JIS, GBK); such edits are now refused.
- Bash permission checks auto-approving some commands that assign the
BASH_ARGV0shell variable and then use it; these now prompt for approval. - Secret redaction in shared transcripts and debug logs missing some values that follow a key with no value, including in JSON written inside a shell string.
- Cloud sessions skipping auto mode’s checks on Claude in Chrome actions that a saved permission allows when their feature flags had not loaded.
- Auto mode occasionally refusing one update to your own artifact, saying its sharing had changed when it had not.
Backgrounding (←) & sessions
- A prompt sent just after
←being run twice, once unseen in the foreground, when the background service was slow to answer. - A prompt sent while SessionStart hooks were still running vanishing when
←moved the session to the background; it is still not sent, but↑now brings it back. --teleportopening an empty conversation when the session’s history could not be read.- Resumed subagents being told that the user rejected a tool call that a session shutdown had interrupted.
CLAUDE_CODE_RESUME_INTERRUPTED_TURNre-running a finished turn after a restart when an MCP tool result had ended that turn.- Toasts and notifications waiting unseen for as long as the
/diffpanel or dialog was open.
Models, MCP & gateway
- Token counts for Haiku 5.5 and other models that take only adaptive thinking, which failed behind some gateways and were counted with budget thinking elsewhere.
- Headless sessions starting a folder’s
.mcp.jsonor plugin MCP server that was switched off for that folder, after changing directory or reloading plugins. - A saved Claude apps gateway sign-in being ignored on machines whose managed settings set
forceLoginMethodtogatewaywith noforceLoginGatewayUrl(regression in 2.1.295). - A Claude apps gateway that serves
allowedProviderswith"gateway"locking out laptops that name that gateway in user settings. - Claude calling cloud sessions non-interactive and suggesting
/mcporclaude mcpwhen an MCP server needs authentication.
Plugins, mods & workflows
- Skills synced from claude.ai never finishing their install, and being downloaded again at every sync, when several sessions share a config directory.
- A plugin synced from claude.ai being disabled when the marketplace dependency it declares was also synced from claude.ai.
- A startup hang on macOS when plugin paths copied from a Linux machine pointed under /home.
claude plugin marketplace add,marketplace updateandplugin installfailing with an internal error for a marketplace named likeconstructor;addnow refuses such names clearly.- A plugin’s served
$methods running in the main session’s working directory instead of the calling agent’s, and ignoring the turn their calling hook holds. - The Workflow tool refusing script files with CRLF line endings, such as files checked out on Windows.
/code-reviewending on a raw JSON array in cloud sessions, the Agent SDK and IDE integrations; the findings now print as a numbered list.
Windows
- PowerShell commands longer than about 1 KB always asking for permission; allow rules and read-only detection now apply up to 32 KB.
- stdio MCP servers being force-killed at shutdown; stdin now closes first and the process tree is killed only if the server is still running after 300 ms.
claude plugin installfailing for GitHubowner/repoplugin sources on machines with no GitHub SSH key; the clone is now retried over HTTPS.rm -rf /c/Users/<name>in Git Bash not asking in bypass permissions mode.
VSCode
claudeCode.spinnerVerbsmissing from the extension’s settings, so settings.json now completes and validates it; a malformed value no longer breaks the chat panel.- Chat links written as a full Windows path, such as C:\repo\file.ts or file:///C:/repo/file.ts, not opening the file on Windows.
- The working indicator still animating when reduced motion is turned on in your OS, in VS Code, or with the
prefersReducedMotionsetting.
Cloud sessions & self-hosted runner
- Every cloud session waiting for a self-hosted runner showing the same queue position; each now shows how many sessions are ahead of it.
- A message you send while Claude is still replying in a cloud session jumping above that reply when it finishes; it now stays below it, including after a reload.
- A self-hosted runner session’s git fetch, pull and push failing while another session of the same repository starts on a runner with
--capacityabove 1. claude self-hosted-runnerprinting misleading errors when registration is refused; it now names the org admin setting, or says a restarted on-demand runner needs a fresh work order.
Claude Tag & Code Review
- [Claude Tag] Links to URLs that contain parentheses breaking in some of Claude’s Slack replies and task checklists, which showed stray bracket text and opened the wrong page.
- [Claude Tag] Claude’s Slack replies drawing color swatches beside codes that only look like hex colors, such as six-digit pull request numbers.
- [Claude Tag] Claude in Slack sometimes losing its reply when a session in a self-hosted environment moved to a new runner partway through a task.
- [Claude Tag] Claude in Slack posting a card to remove a plugin the channel’s own settings don’t include; it now explains where the plugin comes from instead.
- [Code Review] The Code Review check sometimes staying “in progress” on a pull request after the review finished; it now shows the review’s result.
Notes
- If you work with files that are not valid UTF-8 (Windows-1252, Shift-JIS, GBK), Edit and NotebookEdit now refuse those edits instead of replacing every non-ASCII character.
- If a saved Claude apps gateway sign-in was ignored on 2.1.295, that regression, on machines whose managed settings set
forceLoginMethodtogatewaywith noforceLoginGatewayUrl, is fixed here. - Sonnet 5.5 cost figures change.
/cost, the status line and--max-budget-usdnow price cache reads at $0.10 per million tokens (was $0.20). - If you build MCP servers, the default limit on tool descriptions sent up front and on server instructions is now 4,096 characters.
- A turn or
!command started while←moves the session to the background is now stopped instead of finishing out of sight.