Claude docs changes for April 5th, 2026 [diff]

Executive Summary

  • Amazon Bedrock now has an interactive setup wizard accessible from the login screen and via /setup-bedrock
  • New forceRemoteSettingsRefresh managed setting enables fail-closed startup enforcement — the CLI exits rather than starting without managed policies
  • Skills gain multi-line shell execution via fenced ```! blocks, with a new disableSkillShellExecution policy to restrict this
  • /vim command removed in v2.1.92; Vim mode is now toggled via /config → Editor mode
  • Remote Control sessions now support customizable name prefixes via --remote-control-session-name-prefix

Claude Code changes

Changed documents

amazon-bedrock [Source]

  • New interactive Bedrock setup wizard is now available from the login screen (select 3rd-party platformAmazon Bedrock). It guides through AWS authentication, region selection, credential verification, and model pinning. [line 12] [Source]

cli-reference [Source]

  • New --remote-control-session-name-prefix <prefix> flag added. Sets the prefix for auto-generated Remote Control session names (defaults to hostname). Equivalent to CLAUDE_REMOTE_CONTROL_SESSION_NAME_PREFIX. [line 75] [Source]

commands [Source]

  • /pr-comments removed in v2.1.91. Users should ask Claude directly to view pull request comments instead. [line 49] [Source]
  • /release-notes now shows an interactive version picker to select a specific version or view all. [line 51] [Source]
  • New /setup-bedrock command added. Opens the interactive Bedrock configuration wizard (only visible when CLAUDE_CODE_USE_BEDROCK=1). [line 62] [Source]
  • /vim removed in v2.1.92. Vim mode is now toggled via /config → Editor mode. [line 73] [Source]

env-vars [Source]

  • New CLAUDE_REMOTE_CONTROL_SESSION_NAME_PREFIX environment variable documented. Sets the prefix for auto-generated Remote Control session names; defaults to machine hostname. [line 131] [Source]

mcp [Source]

  • Clarified that anthropic/maxResultSizeChars raises the per-tool persist-to-disk threshold but does NOT bypass the global MAX_MCP_OUTPUT_TOKENS limit (~100,000 chars). Users must also raise MAX_MCP_OUTPUT_TOKENS for very large results. [lines 1613-1632] [Source]

permissions [Source]

  • New forceRemoteSettingsRefresh managed-only setting documented. When true, blocks CLI startup until remote managed settings are freshly fetched; CLI exits on fetch failure. [line 245] [Source]

quickstart [Source]

  • Removed the inline install UI snippet (platform/OS picker and install command) from the top of the page, streamlining the quickstart content.

remote-control [Source]

  • New --remote-control-session-name-prefix <prefix> flag documented for the remote-control command. [line 45] [Source]
  • Session title selection order updated: step 4 is now an auto-generated name like myhost-graceful-unicorn (using hostname or configured prefix), with the prompt-based title update noted separately. [lines 92-94] [Source]

server-managed-settings [Source]

  • New "Enforce fail-closed startup" section added, documenting the forceRemoteSettingsRefresh setting. When enabled, the CLI blocks startup until remote settings are fetched and exits on failure. The setting is self-perpetuating via local cache. Requires connectivity to api.anthropic.com. [lines 144-161] [Source]
  • Security model table updated: "API is unavailable" row notes that with forceRemoteSettingsRefresh: true the CLI exits instead of continuing. [line 192] [Source]

settings [Source]

  • New disableSkillShellExecution setting documented. Disables inline shell execution (!`...` and ```! blocks) in skills and custom commands from user/project/plugin sources. Managed settings only; bundled and managed skills are unaffected. [line 167] [Source]
  • New forceRemoteSettingsRefresh setting documented (managed settings only). [line 177] [Source]

skills [Source]

  • shell frontmatter field updated to also apply to fenced ```! code blocks (not just inline !`...`). [line 187] [Source]
  • New multi-line shell execution syntax documented using ```! fenced blocks for running multiple commands in a single block. [lines 363-375] [Source]
  • disableSkillShellExecution policy documented inline — disables shell execution in user/project/plugin skills; commands are replaced with [shell command execution disabled by policy]. [lines 376-377] [Source]

troubleshooting [Source]

  • Added note that sandboxed commands cannot launch Windows binaries (e.g. cmd.exe, powershell.exe, /mnt/c/ executables) due to the WSL Unix socket being blocked by the sandbox. Workaround: add the command to excludedCommands in settings. [line 542] [Source]

API changes

Changed documents

No significant changes today.