Claude docs changes for April 10th, 2026 [diff]

Executive Summary

• New Advisor Tool (Beta): A new API-level server tool lets a faster executor model consult a more powerful advisor model (e.g., Opus) mid-generation for strategic guidance, enabling near-advisor-quality output at lower token cost for agentic workloads.
• Monitor Tool added to Agent SDK: A new Monitor built-in tool lets Claude run a background script and react to each stdout line as an event, replacing polling loops in background task management.
• Claude Code 2.1.98: Major release with a Google Vertex AI interactive setup wizard, claude setup-token for CI authentication, subprocess sandboxing on Linux, Perforce write-protection mode, and dozens of permission/Bash bug fixes.
• New Agent SDK Hooks Guide: A comprehensive new guide explains how to intercept and control agent behavior with hooks in both Python and TypeScript SDKs.
• New Remote MCP Servers: Six new remote MCP servers added (Sybill, Tines, MoSPI, incident.io, Unthread, Zoom); Microsoft 365 connector removed.

New Claude Code versions

2.1.98

New features

• Added interactive Google Vertex AI setup wizard, accessible from the login screen when selecting "3rd-party platform", guiding through GCP authentication, project/region configuration, credential verification, and model pinning
• Added CLAUDE_CODE_PERFORCE_MODE environment variable: when set, Edit/Write/NotebookEdit fail on read-only files with a p4 edit hint instead of silently overwriting them
• Added Monitor tool for streaming events from background scripts (each stdout line delivered to Claude as an event)
• Added subprocess sandboxing with PID namespace isolation on Linux when CLAUDE_CODE_SUBPROCESS_ENV_SCRUB is set; added CLAUDE_CODE_SCRIPT_CAPS env var to limit per-session script invocations
• Added --exclude-dynamic-system-prompt-sections flag to print mode for improved cross-user prompt caching by moving per-machine sections into the first user message
• Added workspace.git_worktree to the status line JSON input, set whenever the current directory is inside a linked git worktree
• Added claude setup-token command to generate long-lived OAuth tokens for CI pipelines and automated environments
• Added W3C TRACEPARENT env var to Bash tool subprocesses when OTEL tracing is enabled, so child-process spans correctly parent to Claude Code's trace tree
• LSP: Claude Code now identifies itself to language servers via clientInfo in the initialize request

Existing feature improvements

• Improved /resume filter hint labels and added project/worktree/branch names in the filter indicator
• Improved /agents with a tabbed layout: a Running tab shows live subagents, and the Library tab adds Run agent and View running instance actions
• Improved /reload-plugins to pick up plugin-provided skills without requiring a restart
• Improved Accept Edits mode to auto-approve filesystem commands prefixed with safe environment variables (e.g., LANG=C) or process wrappers (e.g., timeout, nice)
• Improved Vim mode: j/k in NORMAL mode now navigate history and select the footer pill at the input boundary
• Improved hook errors in the transcript to include the first line of stderr for self-diagnosis without --debug
• Improved OTEL tracing: interaction spans now correctly wrap full turns under concurrent SDK calls

Major bug fixes

• Fixed a Bash tool permission bypass where a backslash-escaped flag could be auto-allowed as read-only and lead to arbitrary code execution
• Fixed compound Bash commands bypassing forced permission prompts for safety checks and explicit ask rules in auto and bypass-permissions modes
• Fixed 429 retries burning all attempts in ~13s when the server returns a small Retry-After — exponential backoff now applies as a minimum
• Fixed MCP OAuth oauth.authServerMetadataUrl config override not being honored on token refresh after restart
• Fixed capital letters being dropped to lowercase on xterm and VS Code integrated terminal when the kitty keyboard protocol is active
• Fixed --dangerously-skip-permissions being silently downgraded to accept-edits mode after approving a write to a protected path via Bash
• Fixed managed-settings allow rules remaining active after an admin removed them, until process restart
• Fixed permissions.additionalDirectories changes not applying mid-session
• Fixed stale subagent worktree cleanup removing worktrees that contain untracked files
• Fixed Bash(cmd:*) and Bash(git commit *) wildcard permission rules failing to match commands with extra spaces or tabs
• Fixed false Bash permission prompts for cut -d /, paste -d /, column -s /, awk '{print $1}' file, and filenames containing %
• Fixed agent team members not inheriting the leader's permission mode when using --dangerously-skip-permissions
• Fixed several /resume picker issues including empty list swallowing arrow keys and transient task-status text replacing conversation summaries
• Fixed /export not honoring absolute paths and ~, and silently rewriting user-supplied extensions to .txt
• Fixed macOS text replacements deleting the trigger word instead of inserting the substitution

Claude Code changes

New Documents

agent-sdk/hooks [Source]

New comprehensive guide for intercepting and controlling agent behavior with hooks in the Agent SDK. Covers all available hook events (PreToolUse, PostToolUse, Stop, SubagentStart/Stop, Notification, etc.), how to configure callback functions, matchers, async outputs, and practical examples including blocking dangerous operations, modifying tool inputs, auto-approving tools, chaining hooks, making HTTP requests, and forwarding notifications to Slack.

Changed documents

agent-sdk/overview [Source]

• Added Monitor tool to the built-in tools table — watches a background script and reacts to each output line as an event. [line ~118]

agent-sdk/python [Source]

• Added full Monitor tool documentation including input schema (command, description, timeout_ms, persistent) and output schema (taskId, timeoutMs, persistent).
• Updated TaskStartedMessage and TaskNotificationMessage to mention Monitor watches as a background task type.

agent-sdk/typescript [Source]

• Added Monitor tool with MonitorInput and MonitorOutput types to the TypeScript SDK reference.
• Updated SDKTaskStartedMessage and SDKTaskNotificationMessage to note that Monitor watches use task_type: "local_bash".

amazon-bedrock [Source]

• Improved "Sign in with Bedrock" section with a new step-by-step wizard flow (wizard detects credentials, region, available models, and writes settings automatically).
• Restructured setup into "Sign in with Bedrock" (for individuals) vs "Set up manually" (for CI/enterprise rollouts).
• Added new "Startup model checks" section: Claude Code verifies model accessibility at startup and can fall back to the previous version when the latest is unavailable in Bedrock.

authentication [Source]

• Added CLAUDE_CODE_OAUTH_TOKEN as a new credential priority tier (between apiKeyHelper and subscription OAuth).
• Added new "Generate a long-lived token" section documenting claude setup-token for CI environments — generates a one-year OAuth token without saving it.

cli-reference [Source]

• Added claude setup-token command for generating long-lived OAuth tokens.
• Added --exclude-dynamic-system-prompt-sections flag that moves per-machine sections to the first user message for better prompt-cache reuse in multi-user workloads.

commands [Source]

• Renamed from "Built-in commands" to "Commands" and updated description to distinguish built-in commands from bundled skills.
• Added /batch skill for orchestrating large-scale parallel codebase changes.
• Added /claude-api skill for loading Claude API reference material.
• Added /debug skill for enabling debug logging and troubleshooting mid-session.
• Added /loop skill for running prompts on a recurring interval.
• Added /setup-vertex command for configuring Google Vertex AI interactively.
• Added /simplify skill for reviewing changed code for quality and efficiency.

common-workflows [Source]

• Subagent worktree cleanup now requires no uncommitted changes, no untracked files, and no unpushed commits (previously only checked modifications to tracked files).

context-window [Source]

• Added new "What survives compaction" section with a detailed table showing what happens to each context mechanism (system prompt, CLAUDE.md, skills, hooks, path-scoped rules) after /compact.

costs [Source]

• Added note that for organizations with custom rate limits, Claude Code traffic counts toward API rate limits, and a workspace rate limit can be set in the Claude Console to cap Claude Code's share.

env-vars [Source]

• Added CLAUDE_CODE_PERFORCE_MODE for Perforce-aware write protection on read-only files.
• Added CLAUDE_CODE_SCRIPT_CAPS for limiting per-session script invocations by command substring.
• Updated CLAUDE_CODE_SUBPROCESS_ENV_SCRUB to document new Linux PID namespace isolation behavior.
• Updated CLAUDE_CODE_OAUTH_TOKEN to reference claude setup-token for generation.

features-overview [Source]

• Added new "Build your setup over time" section with a trigger table guiding when to add CLAUDE.md, skills, MCP servers, subagents, hooks, or plugins.

fullscreen [Source]

• Added note that scroll actions are rebindable and a reference to the Scroll actions section in keybindings.

google-vertex-ai [Source]

• Added new "Sign in with Vertex AI" section with a step-by-step wizard flow requiring v2.1.98 or later.
• Restructured into "Sign in" (individuals) vs "Set up manually" (CI/enterprise).
• Added new "Startup model checks" section: Claude Code verifies model accessibility at startup with fallback and pin-update prompting.

hooks [Source]

• Updated hook lifecycle description to clarify event cadences: once per session, once per turn, and per tool call.
• Improved matcher pattern documentation with a table distinguishing exact-string matchers (Edit|Write) from regex matchers (containing special characters).
• Hook error display now shows the first line of stderr in the transcript (not just the hook name), aiding self-diagnosis without --debug.
• Clarified FileChanged matcher behavior: watch list is built from literal filenames split on |, not regex patterns.
• MCP tool matcher guidance updated: mcp__memory (letters/underscores only) matches as exact string, so .* suffix is required for prefix matching.

hooks-guide [Source]

• Updated exit code documentation: non-blocking errors now display hook name plus first line of stderr in transcript.
• Updated FileChanged matcher description to clarify literal filename watch list vs regex filtering.

index [Source]

• Added documentation for two Homebrew casks: claude-code (stable channel, ~1 week behind) and claude-code@latest (latest channel, immediate updates).

interactive-mode [Source]

• Updated Vim mode documentation: j/k now navigate history in NORMAL mode, in addition to arrow keys.

keybindings [Source]

• Added new Scroll keybinding context for fullscreen mode with 12 scroll and selection actions (line up/down, page up/down, half/full page, top, bottom, copy selection, clear selection).

mcp [Source]

• Added Sybill MCP server (sales calls and pipeline AI).
• Removed Microsoft 365 MCP connector.
• Updated Slack MCP connection command to include --client-id and --callback-port parameters.

memory [Source]

• Added "When to add to CLAUDE.md" guidance with specific triggers (repeated mistakes, code review findings, recurring corrections).
• Updated compaction behavior: nested CLAUDE.md files in subdirectories are not re-injected automatically after /compact; only the project-root file is.

model-config [Source]

• Updated third-party deployment pin documentation: Bedrock and Vertex AI now fall back to the previous model version at startup when the latest is unavailable (Foundry still errors without pinning).

monitoring-usage [Source]

• Updated attempt attribute definition: now represents total attempts made (1 = no retries).
• Added new "Detect retry exhaustion" section explaining how to use attempt to distinguish transient errors from non-retryable ones.

overview [Source]

• Added documentation for two Homebrew casks: claude-code (stable) and claude-code@latest (latest).

permission-modes [Source]

• Updated acceptEdits mode: now auto-approves filesystem commands prefixed with safe env vars (LANG, NO_COLOR, etc.) or process wrappers (timeout, nice, nohup).

API changes

New Documents

agents-and-tools/tool-use/advisor-tool [Source]

New documentation for the Advisor tool (beta: advisor-tool-2026-03-01). The advisor tool lets a faster executor model consult a more powerful advisor model mid-generation for strategic guidance. The advisor reads the full conversation transcript and returns a plan (typically 400–700 text tokens); the executor continues with the task at its lower rate. Documents model compatibility table (Haiku/Sonnet/Opus as executors, Opus as advisor), the advisor_20260301 tool type, response structure with advisor_result and advisor_redacted_result variants, multi-turn usage, streaming behavior, usage and billing via usage.iterations, two-layer prompt caching (executor-side and advisor-side), combining with other tools, best practices for system prompt design, and cost control.

Changed documents

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

• Added Sybill MCP server (https://mcp.sybill.ai/mcp) for sales calls, deals, and pipeline AI.
• Added Tines MCP server (user-specific URL) for workflow automation.
• Added MoSPI MCP server (https://mcp.mospi.gov.in/) for India's official statistics.
• Added incident.io MCP server (https://mcp.incident.io/mcp) for incident management.
• Added Unthread MCP server (https://app.unthread.io/api/mcp) for support ticket automation.
• Added Zoom for Claude MCP server (https://mcp.zoom.us/mcp/zoom/streamable) for meeting search and recaps.
• Removed Microsoft 365 MCP connector.
• Updated Aiwyn Tax description.

agents-and-tools/tool-use/tool-reference [Source]

• Added Advisor tool (advisor_20260301) to the server tools table with Beta status.

api/beta [Source]

• Added advisor-tool-2026-03-01 to the list of valid AnthropicBeta values.
• Added BetaAdvisorMessageIterationUsage model for advisor sub-inference token usage tracking.

api/beta/messages/create [Source]

• Added advisor-tool-2026-03-01 beta header option.
• Added advisor to server tool names and documented BetaAdvisorToolResultBlockParam with advisor_tool_result_error, advisor_result, and advisor_redacted_result content variants.