Claude Code v2.1.271: Fast Mode in Remote Sessions, Per-Command Sandbox Domains, and omitClaudeMd

Claude Code v2.1.271 is a large release with 96 changelog entries. It brings fast mode to Claude Code Remote sessions, adds per-command network allowances for sandboxed commands in auto mode and an omitClaudeMd option for subagents, and changes how auto mode and Monitor watches behave. Many of the fixes target organization policy loading, Bash permission checks, MCP and session resume.

What’s New

Fast mode in Claude Code Remote sessions

Fast mode now works in Claude Code Remote sessions, on both cloud and self-hosted runners. The host’s fast-mode setting, or /fast typed in the session, applies where your organization allows it.

Per-command allowed domains in sandboxed auto mode

In auto mode with sandboxing, Bash, PowerShell and Monitor now support per-command allowed_domains. The hosts a command needs are reviewed together with the command and opened for that command alone; other hosts are refused.

`omitClaudeMd` for subagents

Agent frontmatter and --agents JSON now accept omitClaudeMd, which lets custom and plugin subagents run without user, project and local CLAUDE.md files. Managed policy files still load.

Plugin command acceptance and pricing multipliers

claude plugin install and claude plugin update gain --accept-command <sha256>, which accepts exactly the command a previous --json run displayed, instead of -y. The modelPricing managed setting and the Claude apps gateway pricing block now accept a multiplier above 1, up to 10, for marked-up internal chargeback rates.

Self-hosted runner options

With claude self-hosted-runner --drain-marker-file <path>, when that file exists at a SIGTERM drain, the runner reports its exit to the server as a host drain (telemetry only). A new --host-config-snapshot disk|memory option arrives with a fix for runner sessions that silently lost all host config (settings, skills, plugins, MCP servers) when the host config directory exceeded 64 MiB.

Mouse support in `/config`

In fullscreen mode, the /config panel now supports the mouse: the wheel scrolls the settings list, clicking a setting’s value changes it, and the row under the pointer is highlighted.

Auto mode and Monitor behavior changes

In auto mode, a skill’s or slash command’s inline ! shell commands now follow default-mode permission rules instead of the classifier; a command that no rule decides runs as a reviewed tool call.
In auto mode, a subagent now reports back to its caller through a dedicated hand-back call that the safety classifier reviews, instead of its last message being reviewed after the fact.
Monitor watches now always have a deadline (at most 30 minutes; 10 in single-prompt -p runs) and notify Claude to re-arm, replacing the no-timeout persistent option.

Dynamic workflow changes

Dynamic workflows now pause when you hit your usage limit and continue automatically when it resets, instead of dropping the affected agents. The default dynamic workflow size on Pro plans is now small, and the medium size guideline is lowered from 15 to 10 agents.

Fixes and Improvements

Organization policy: A cached organization policy is no longer reused after switching accounts, organizations or API keys, and it now refreshes when the credential changes mid-session instead of waiting for the hourly check. Tool and command lists also update when the policy finishes loading after startup or changes mid-session.
Managed MCP: An enterprise managed-mcp.json that can’t be read or parsed is no longer ignored; it now keeps exclusive MCP control (user, project and plugin servers don’t load) and warns at startup.
Bash permission checks: Checks no longer miss the file that fmt, column and similar commands read after an option the checker doesn’t recognize, no longer skip files a wildcard in a pattern or option value expands to (for example grep -v dir/* file), and can no longer be misled by shell variable declaration flags. Commands with two directory changes, a subshell, or a cd+git chain no longer skip the prompt under permissions.blockReadsOutsideWorkingDirectories in bypass and auto mode.
Fast mode: /fast off now turns fast mode off instead of answering “Fast mode unavailable” when the organization has fast mode disabled; sessions started with CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK no longer re-send fast requests every turn after the API rejects fast mode; and under CLAUDE_CODE_RETRY_WATCHDOG, fast mode now falls back to standard speed on a usage-credits limit or an overload.
Resume: /resume and /teleport no longer keep the previous conversation’s file-read tracking, which could let Claude edit files the resumed conversation had never read. --resume no longer drops the 1M context window ([1m]) when the resumed session’s model family differs from the configured default model, and artifacts attached with /artifacts no longer disappear after --resume.
Background commands: Claude no longer starts a second copy of a background command, such as a watch task or dev server, that was still running after the conversation was compacted.
Git on Linux: A stale .git/config.lock left after a sandboxed command failed to start no longer breaks git checkout -b, git push -u and git config for the rest of the session.
Gateways and proxies: Turns no longer fail with “API returned an empty or malformed response” when an LLM gateway returns the non-streaming reply as text/plain, and third-party local proxies set via ANTHROPIC_UNIX_SOCKET are again treated like other custom gateways, including for Remote Control. Claude apps gateway, Bedrock, Vertex AI and Foundry sessions no longer refresh a leftover claude.ai login they don’t use.
MCP: Fixed sustained high CPU usage when an MCP server sends list_changed notifications in a tight loop, several MCP OAuth client-registration problems, and tool search returning no match when Claude selects an MCP tool by its bare name. claude mcp serve now sends a progress update every 30 seconds during a running tool call, so idle timeouts don’t abort a long command that prints nothing.
Hooks: While a SessionStart, UserPromptSubmit, PreToolUse or SessionEnd hook runs, the spinner now says so with elapsed time, and Esc cancels a prompt waiting on a SessionStart hook. The interactive /hooks menu no longer crashes when a matcher is named after an inherited object property such as __proto__ or constructor.
Skills: Skills synced from claude.ai no longer stay on disk indefinitely after signing out; copies not refreshed within cleanupPeriodDays move to the recoverable trash at the next launch. The bundled claude-api skill now enables eager_input_streaming on streaming custom tools and starts deliverable-shaped Managed Agents work with user.define_outcome.
Artifacts: Markdown files published as artifacts now render as styled document pages, a session can watch up to 10 published artifacts at once for republishes made elsewhere (up from 5), and background sessions (claude --bg, claude agents) now watch the artifacts they publish.
Prompt input: The IDE selection indicator is now a [⧉ …] pill that wraps with the text instead of squeezing multi-line prompts; delete it with Backspace to leave the selection out. The /add-dir path input now moves the cursor with the arrow keys, and Enter adds only the typed path.
Performance: Large diffs and long transcripts render faster with fewer slow frames, and startup is slightly faster because a redundant validation of built-in model data is skipped.
Windows: PowerShell commands no longer fail with “Exit code 1” and no output when the session’s temp output path reaches 260 characters, and custom agents, slash commands and output styles beyond the first now load from a virtual drive that reports inode 0.
VS Code: A new Attach Open File setting, when turned off, stops the open file from being added to messages (selected text is still attached). The Hooks and Permission rules dialogs received several save fixes, and console windows no longer flash on Windows when the extension runs background commands.
Claude Code on the web: A cloud session no longer sometimes takes about ten minutes to respond after its process exited, the Routines page has a new layout with Yours and Templates tabs (its calendar view was removed), and the Cloud environments editor in admin settings gains a Custom network access option.
Claude Tag: Claude in a channel where it stays active no longer loses its working context about once an hour when the conversation is mostly in threads, and only instructions addressed to Claude now bind it; it asks when unsure.
Code Review: Fixed the same findings occasionally being posted two or three times, follow-up reviews re-posting a security finding a person had already resolved, and reopening a finished /ultrareview cloud session restarting the whole review.

This release contains 96 changes in total; see the full changelog for the complete list.

How to Update

Check your installed version: claude --version
Update in place: claude update
If you use Monitor watches with the persistent option, note that watches now always have a deadline (at most 30 minutes; 10 in -p runs) and Claude is notified to re-arm them. In auto mode, inline ! shell commands in skills and slash commands now follow your default-mode permission rules.