Claude docs changes for April 14th, 2026 [diff]

Executive Summary

• Version 2.1.105 is a major release: PreCompact hooks can now block compaction, a new asyncRewake hook type wakes Claude on background failures, and skills get a new when_to_use field with a raised 1,536-character description cap
• Bash permission rules gained a detailed "Process wrappers" section documenting which wrappers (timeout, nice, nohup, xargs, etc.) are stripped before matching, and deny/ask rules are now evaluated regardless of hook return values
• subagentStatusLine is a new setting that lets you render custom rows in the agent panel, and plugins can now ship monitors that auto-arm at session start
• New telemetry events added: plugin_installed and skill_activated, and OTEL_LOG_TOOL_DETAILS now also populates trace span attributes
• Two new MCP servers listed: Kindora Funder Discovery and AdisInsight (pharmaceutical intelligence)

New Claude Code versions

2.1.105

New features

• Added path parameter to the EnterWorktree tool to switch into an existing worktree of the current repository
• Added PreCompact hook support: hooks can now block compaction by exiting with code 2 or returning {"decision":"block"}
• Added background monitor support for plugins via a top-level monitors manifest key that auto-arms at session start or on skill invoke
• /proactive is now an alias for /loop

Existing feature improvements

• Improved stalled API stream handling: streams now abort after 5 minutes of no data and retry non-streaming instead of hanging indefinitely
• Improved network error messages: connection errors now show a retry message immediately instead of a silent spinner
• Improved file write display: long single-line writes are now truncated in the UI
• Improved /doctor layout with status icons; press f to have Claude fix reported issues
• Improved /config labels and descriptions for clarity
• Improved skill description handling: raised the listing cap from 250 to 1,536 characters and added a startup warning when descriptions are truncated
• Improved WebFetch to strip <style> and <script> contents from fetched pages
• Improved stale agent worktree cleanup to remove worktrees whose PR was squash-merged
• Improved MCP large-output truncation prompt to give format-specific recipes

Major bug fixes

• Fixed images attached to queued messages being dropped
• Fixed screen going blank when the prompt input wraps to a second line in long conversations
• Fixed leading whitespace getting copied when selecting multi-line assistant responses in fullscreen mode
• Fixed garbled bash output when commands print clickable file links
• Fixed alt+enter not inserting a newline in terminals using ESC-prefix alt encoding, and Ctrl+J not inserting a newline (regression in 2.1.100)
• Fixed queued user prompts disappearing from focus mode
• Fixed one-shot scheduled tasks re-firing repeatedly
• Fixed inbound channel notifications being silently dropped after the first message for Team/Enterprise users
• Fixed marketplace plugins with package.json and lockfile not having dependencies installed automatically after install/update
• Fixed stdio MCP server emitting malformed output hanging the session instead of failing fast
• Fixed MCP tools missing on the first turn of headless/remote-trigger sessions
• Fixed /model picker on AWS Bedrock in non-US regions persisting invalid model IDs
• Fixed 429 rate-limit errors showing a raw JSON dump instead of a clean message
• Fixed crash on resume when session contains malformed text blocks
• Fixed CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC in one project permanently disabling usage metrics for all projects
• Fixed washed-out 16-color palette when using Ghostty, Kitty, Alacritty, WezTerm, foot, rio, or Contour over SSH/mosh
• Fixed Bash tool suggesting acceptEdits permission mode when exiting plan mode would downgrade from a higher permission level

2.1.107

Existing feature improvements

• Show thinking hints sooner during long operations

Claude Code changes

New Documents

whats-new/2026-w15 [Source]

Week 15 (April 6–10, 2026) digest covering releases v2.1.92–v2.1.101. Highlights: Ultraplan enters early preview (plan in the cloud from CLI, review in browser, execute remotely or locally); Monitor tool streams background events into the conversation for live log tailing and reactions; /loop self-pacing when interval is omitted; /team-onboarding generates a teammate ramp-up guide; /autofix-pr enables PR auto-fix from the terminal.

Changed documents

agent-sdk/typescript [Source]

• EnterWorktreeInput now has an optional path field to switch into an existing worktree; name and path are mutually exclusive. [line ~1597]

commands [Source]

• /doctor description updated: results now show with status icons and pressing f sends the report to Claude for fixing.
• /loop now lists /proactive as an alias.
• New /team-onboarding command: generates a team onboarding guide from the last 30 days of Claude Code usage history.

common-workflows [Source]

• claude --resume now accepts custom session names (set with --name or /rename) in addition to session IDs.

costs [Source]

• Added a reference to claude.com/pricing for subscription plan pricing (Pro, Max, Team, Enterprise).

discover-plugins [Source]

• Added troubleshooting guidance for when a plugin is not found: run /plugin marketplace update claude-plugins-official to refresh, or /plugin marketplace add anthropics/claude-plugins-official to add it.

env-vars [Source]

• New ANTHROPIC_CUSTOM_MODEL_OPTION_SUPPORTED_CAPABILITIES env var for specifying capabilities of custom model entries.
• CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD updated: now also loads CLAUDE.local.md from additional directories (skipped if local is excluded from --setting-sources).
• CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MS updated: default budget is now auto-raised to the highest per-hook timeout configured in settings files, up to 60 seconds.
• New CLAUDE_ENABLE_BYTE_WATCHDOG env var: controls a byte-level streaming watchdog (enabled by default for Anthropic API) that aborts when no bytes arrive for the duration set by CLAUDE_STREAM_IDLE_TIMEOUT_MS (minimum 5 minutes).
• CLAUDE_ENABLE_STREAM_WATCHDOG now only enables the event-level watchdog (off by default); the byte-level watchdog is the new default mechanism for Anthropic API.
• CLAUDE_STREAM_IDLE_TIMEOUT_MS updated with per-watchdog defaults: 5 minutes (minimum) for byte-level, 90 seconds for event-level.
• OTEL_LOG_TOOL_DETAILS updated: now also populates tool input in trace span attributes in addition to log events.
• New VERTEX_REGION_CLAUDE_4_5_OPUS and VERTEX_REGION_CLAUDE_4_6_OPUS region override env vars for Vertex AI.

hooks [Source]

• New asyncRewake command hook field: runs in the background and wakes Claude immediately when it exits with code 2, delivering the hook's stderr as a system reminder. [line ~291]
• PreCompact hooks can now block compaction via exit code 2 or {"decision":"block"} JSON; previously they were side-effect only. [line ~528]
• Documented the two different effects of blocking PreCompact: skips proactive compaction vs. surfaces context-limit error for recovery compaction. [line ~1788]
• SessionEnd hooks timeout behavior updated: per-hook timeout settings automatically raise the overall budget up to 60 seconds; the env var is now an explicit override rather than an absolute cap. [line ~1863]
• Clarified that PreToolUse hook permissionDecision values do not bypass deny/ask rules — those are evaluated regardless. [line ~930]

interactive-mode [Source]

• Ctrl+O description updated: MCP calls now collapse to "Called slack 3 times" style (previously only read and search calls collapsed).

keybindings [Source]

• New Doctor keybinding context with a doctor:fix action (default key: F) to send diagnostics to Claude for fixing.
• chat:newline now has a default binding: Ctrl+J.

mcp [Source]

• New MCP server added: Kindora Funder Discovery — find funders who support causes like yours.
• New MCP server added: AdisInsight — pharmaceutical drug & clinical trial intelligence.
• Updated links for bioRxiv, ChEMBL, Clinical Trials, CMS Coverage, ICD-10 Codes, and NPI Registry servers.
• Asana transport updated from streamable-http to http.

memory [Source]

• CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD now also loads CLAUDE.local.md from additional directories; previously only CLAUDE.md, .claude/CLAUDE.md, and .claude/rules/*.md were loaded.

model-config [Source]

• ANTHROPIC_CUSTOM_MODEL_OPTION now supports the _NAME, _DESCRIPTION, and _SUPPORTED_CAPABILITIES suffixes, matching the pattern already available for the default Opus/Sonnet/Haiku model env vars.

monitoring-usage [Source]

• New claude_code.plugin_installed telemetry event with attributes including plugin name, version, marketplace source, and install trigger (cli or ui).
• New claude_code.skill_activated telemetry event with attributes including skill name, source, and owning plugin.
• OTEL_LOG_TOOL_DETAILS now also includes tool_input and derived attributes (e.g., file_path) in trace spans, not just log events.

network-config [Source]

• Chrome integration requires bridge.claudeusercontent.com to be allowlisted for outbound WebSocket connections.

permissions [Source]

• Clarified that the :* trailing wildcard is only recognized at the end of a pattern; in the middle (e.g., Bash(git:* push)) the colon is treated literally.
• A single * matches any sequence including spaces, so one wildcard spans multiple arguments.
• New "Compound commands" section: documented recognized command separators (&&, ||, ;, |, |&, &, newlines) and that each subcommand must match independently.
• New "Process wrappers" section: timeout, time, nice, nohup, stdbuf, and bare xargs are stripped before matching rules. Development environment runners (devbox, mise, npx, docker exec, etc.) are NOT stripped — write rules that include the runner.
• Hook decisions clarified: deny/ask permission rules are evaluated regardless of what a PreToolUse hook returns — a deny rule blocks even when the hook returned "allow".
• CLAUDE.local.md is now also loaded from --add-dir directories (requires CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1 and the local setting source).

plugins [Source]

• Plugin settings.json now supports the subagentStatusLine key in addition to agent.

plugins-reference [Source]

• New monitors manifest key: background Monitor configurations that auto-arm when the plugin is enabled at session start or when a skill in the plugin is invoked.
• Plugin directory structure updated to include monitors/monitors.json as a recognized location.
• Plugin settings.json now supports subagentStatusLine in addition to agent.

settings [Source]

• New viewMode setting: sets the default transcript view mode on startup ("default", "verbose", or "focus"), overriding the sticky Ctrl+O selection.

setup [Source]

• Added a comparison table for native Windows vs WSL 2 vs WSL 1, including sandboxing support.
• Expanded Windows setup documentation with clearer guidance on PowerShell vs CMD and how Claude Code uses Git Bash internally.
• Clarified that WSL setups do not require Git for Windows.

skills [Source]

• New "Live change detection" section: skill directories are watched for changes; edits take effect within the current session without restarting.
• New when_to_use frontmatter field for skills: additional context on when Claude should invoke the skill, appended to description in the listing.
• Skill description cap raised from 250 to 1,536 characters, now covering the combined description + when_to_use text.

statusline [Source]

• New subagentStatusLine setting: renders custom row bodies in the agent panel for each subagent. The command receives subagent task data via stdin and writes one JSON line per row to override. [line ~569]

sub-agents [Source]

• Clarified that frontmatter hooks in subagent files only fire when the agent is spawned via the Agent tool or an @-mention, not when it runs as the main session via --agent or the agent setting.

tools-reference [Source]

• EnterWorktree tool description updated: pass a path to switch into an existing worktree instead of creating a new one.
• Monitor tool: now also unavailable when DISABLE_TELEMETRY or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC is set.

ultraplan [Source]

• Ultraplan now automatically creates a default cloud environment on first launch if one doesn't exist.

API changes

Changed documents

agents-and-tools/agent-skills/quickstart [Source]

• Code example tabs expanded to include Python and TypeScript variants across all examples.

agents-and-tools/mcp-connector [Source]

• Code example tabs expanded to include more language variants.

agents-and-tools/remote-mcp-servers [Source]

• New MCP server listed: Kindora Funder Discovery — find funders who support causes like yours (https://kindora-mcp.azurewebsites.net/mcp/).
• New MCP server listed: AdisInsight — pharmaceutical drug & clinical trial intelligence (https://adisinsight-mcp.springer.com/mcp).
• Updated link for ICD-10 Codes to point to the official Claude tutorial page.

build-with-claude/adaptive-thinking [Source]

• Fixed contact sales link for requesting full thinking output for Claude 4 models (now uses a plain mailto: link).
• Code example tabs expanded to include more language variants.

managed-agents/agent-setup [Source]

• Code example tabs expanded to include curl, Python, TypeScript, and other language variants.

managed-agents/environments [Source]

• Code example tabs expanded to include more language variants across all examples.

managed-agents/memory [Source]

• Code example tabs expanded to include more language variants across all examples.

managed-agents/sessions [Source]

• Code example tabs expanded to include more language variants across all examples.