claudekit / updates / claude-code-2-1-302
[ PATCH · ]

Claude Code 2.1.302

Adds `CLAUDE_CODE_AUTO_MODE_MODEL` to choose the model the client-side auto mode classifier uses, an `Idle compaction` row in `/config`, and a notice that standalone Claude Design closes by December 14, 2026. Fork subagents are now on by default in `claude -p`, the Agent SDK, the desktop app and VS Code, and startup is faster with many plugins enabled. Fixes a "Prompt is too long" error when stopping a turn during compaction, and the Read tool loading image files of any size into memory.

Official announcement →

This article is a summary based on official documentation.

What changed

Claude Code 2.1.302 shipped on October 16, 2026. It adds a model setting for the auto mode classifier, a way to turn idle compaction off, and a banner explaining why a cloud routine didn’t start. Fork subagents and how subagents run are now consistent between the terminal and other surfaces. The fixes span compaction and the prompt cache, provider switching, the Read and PowerShell tools, plugin settings, MCP, OpenTelemetry, terminal rendering, cloud sessions, Claude Tag in Slack and the VS Code extension. The release is large, so the items below are a selection of the changes with the most user impact.

New features

  • CLAUDE_CODE_AUTO_MODE_MODEL

    There was no way to choose which model the client-side auto mode classifier uses. CLAUDE_CODE_AUTO_MODE_MODEL now sets it. It does not change the model for server-side auto mode.

  • Idle compaction row in /config

    /config now has an Idle compaction row for turning idle compaction off or back on, and the notices shown after an idle compaction now say how to turn it off.

  • Standalone Claude Design closing notice

    /design-sync and /design-login now show a notice that standalone Claude Design (claude.ai/design) closes by December 14, 2026.

  • [Cloud sessions] Why a routine’s last run didn’t start

    A routine’s page now has a banner that says why its last run didn’t start, such as your organization reaching its limit on concurrent cloud sessions.

Key improvements

  • Startup time with many plugins enabled

    Plugin command and agent files now take far fewer file system calls to load, most of all on Windows.

  • Auto mode with the server-side classifier

    A conversation too long for the safety check used to get every checked tool call denied. It is now compacted instead.

  • Clearer error messages

    A background session reopened from another folder with no LLM gateway now says why, instead of “Not logged in”. A cloud session that fails to start now says “Connection lost” or “Server error” instead of “Something went wrong”.

Changed behavior

  • Fork subagents on by default

    Fork subagents (subagent_type: "fork") are now 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.

  • Subagents in the desktop app and VS Code

    The desktop app and VS Code now always run the main conversation’s subagents in the background, as the terminal does, rather than letting Claude choose. A subagent’s own Agent calls in claude -p, the Agent SDK, the desktop app and VS Code now run in the foreground by default.

  • Default permission mode on third-party providers

    claude -p and Agent SDK sessions with no permission mode now start in default mode on third-party providers, with telemetry off or on a fresh install. --permission-mode auto keeps auto.

  • Concurrency limit for prompt and agent hooks

    At most 12 prompt or agent hooks of one event run at a time; the rest wait their turn.

  • Claude in Chrome instructions in the system prompt

    The Claude in Chrome instructions moved within the system prompt so that new conversations can read them from the shared prompt cache.

  • Other display changes

    The ← for agents footer hint now hides 10 seconds after a session opens; the count of agents waiting on you still shows. When your organization’s policy turns off /feedback, the message now points to GitHub issues instead of an organization admin.

Bug fixes

Compaction & prompt cache

  • A “Prompt is too long” error when stopping a turn while the conversation was being compacted.
  • A prompt cache miss when a tool added mid-session was loaded again through ToolSearch or after compaction.

Models & providers

  • Requests failing with “diagnostics: Extra inputs are not permitted” after switching to Vertex or Bedrock mid-session.
  • /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.
  • Prompt suggestions spending usage credits on models your plan pays for only with credits; they are now skipped after a reply billed to credits.

Tools & worktrees

  • The Read tool loading an image file of any size, or a pipe or device named like an image, entirely into memory; image sources over 128 MB are now refused.
  • The PowerShell tool not finding commands in a plugin’s 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.
  • Worktree isolation refusing to create or resume worktrees under git 2.56 or later when a stray .git file points at a directory that doesn’t exist.

Plugins & settings

  • 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.
  • Plugin marketplaces re-cloning on every refresh, and newly installed git plugins showing version “unknown”, when git defaults to the reftable ref format.
  • claude plugin details <name>@builtin suggesting claude plugin list when the plugin isn’t found, though that never shows built-in plugins; it now points to /plugin.

MCP & OpenTelemetry

  • MCP servers showing no tools, or every tool call failing, when the server negotiates protocol version 2026-07-28 but omits resultType, ttlMs or cacheScope.
  • Claude in the desktop app pointing you to claude.ai settings or /mcp to authorize or reconnect an MCP server where that option isn’t available.
  • 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.

Artifacts & Claude Design

  • Artifact publishes sometimes failing when Claude sent an empty list of edits, and Claude retrying an artifact update the wrong way after a newer version of the artifact turned it away.
  • /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.

Terminal display

  • Rows in the agents view staying blank after moving down a long list or folding a section.
  • URLs written in backticks being clickable only on their first row when they wrap; the whole URL is now one link in terminals that support hyperlinks.
  • An emoji that a link starts inside wrapping one column too wide, which hid it in question previews and turned tables into lists.
  • claude remote-control --help not printing its usage and exiting when not signed in.

Cloud sessions & Remote Control

  • “Rewind to here” failing on every message except the newest in cloud sessions; self-hosted runners need Claude Code v2.1.246 or later.
  • A scheduled routine run that failed partway being run again from the start an hour or more later, which could repeat work such as posting the same report twice.
  • Routine runs showing as Completed after their cloud session stopped unexpectedly, and as Cancelled when archived before they started, instead of failed.
  • A cloud session’s cost and usage restarting from zero after the session resumed in a new process.
  • The diff view missing changes or failing to load after the session’s GitHub repository was renamed or moved.
  • The pull request status bar flipping a pull request with merge conflicts back to ready while GitHub rechecked it.
  • An open cloud session sometimes not showing messages sent from another device until you reloaded the page or clicked away and back.
  • [Remote Control] A subagent’s pane spinning forever after your computer goes offline; it now says Remote Control is disconnected.

Claude Tag (Slack)

  • HEIC, SVG, TIFF and BMP images attached in Slack never reaching Claude; they now arrive as files in its working files, like other file types.
  • New channel managers losing access to Claude in Slack in organizations that restrict Claude Tag by role; the manager role now also grants Claude Tag in Slack.
  • Claude in Slack sometimes answering a direct message with “DMs are temporarily unavailable” and dropping it when a single Slack lookup stalled; it now retries first.
  • Direct messages on a Slack Enterprise Grid ignoring Slack’s workspace dropdown; they now run under the connected workspace you last opened Claude in.
  • URLs in code blocks under a numbered or bulleted step being wrapped in angle brackets, which broke copied commands, and links absorbing the closing bold or italic marks, or full-width punctuation, right after them.
  • “is thinking…” staying up under a mention after Claude answered it by editing one of its earlier messages.
  • Custom connectors whose MCP servers share one host showing as a single connector in Claude Tag admin settings, and the monthly spend limit notice saying a channel manager can raise it; it now names an organization owner and links the usage page.

Claude Tag admin settings also changed: Claude’s access is now organized into four tabs (Bundles, Connectors, Skills and plugins, and Channels), and repository search covers up to 100 connected GitHub accounts and finds private repositories past the first 5,000 of a GitHub App install.

VSCode

  • Text after an unclosed code block in a list showing as plain text while a reply streams, then turning into code.
  • MCP server sign-in from the panel opening a broken authorization page when the server URL contains a query string.
  • The @browser banner saying “Browser connected” when no browser with the Claude extension can be reached.

Windows, self-hosted runner & Code Review

  • [Windows] Every file write being refused or asked about for projects on exFAT or FAT32 drives, including in sessions started with --plugin-dir.
  • [Self-hosted runner] git fetches failing with “bad object refs/heads” in sessions that reuse a checkout stored in git’s reftable format.
  • [Code Review] Code Review setup not resuming after you leave to connect GitHub; back in admin settings, it now reopens at the repository picker.

Notes

  • If you run claude -p or the Agent SDK on a third-party provider, sessions with no permission mode may now start in default mode. Pass --permission-mode auto to keep auto.
  • If you don’t want fork subagents in claude -p, the Agent SDK, the desktop app or VS Code, set CLAUDE_CODE_FORK_SUBAGENT=0.
  • If you attach many prompt or agent hooks to one event, at most 12 now run at a time and the rest wait their turn.
  • If you don’t want idle compaction, turn it off with the Idle compaction row in /config.
  • If you use standalone Claude Design (claude.ai/design), it closes by December 14, 2026.