Updates

Added `terminalSequence` to hook JSON output so hooks can emit desktop notifications, window titles, and bells without a controlling terminal, `CLAUDE_CODE_PLUGIN_PREFER_HTTPS` to clone GitHub plugin sources over HTTPS instead of SSH, `ANTHROPIC_WORKSPACE_ID` for workload identity federation, and `claude agents --cwd <path>` to scope sessions to a directory. `/feedback` can now include recent sessions (last 24 hours or 7 days), and the Rewind menu adds "Summarize up to here" to compress earlier context. Bug fixes span background side-queries sending an unavailable Haiku model ID on Bedrock/Vertex/Foundry/gateway, Remote Control MCP connectors failing with 401 during worker token rotation, `claude plugin install` failing on stale marketplace refs, MCP HTTP/SSE 403s being shown as `failed` instead of `needs auth`, and a wide sweep of auth, session, MCP, plugin, and UI regressions.

Agent tool `subagent_type` matching now accepts case- and separator-insensitive values (e.g. "Code Reviewer" resolves to `code-reviewer`), and plugins now warn when a default component folder (e.g. `commands/`) is silently ignored because `plugin.json` sets the matching key. Bug fixes cover `/goal` silently hanging under `disableAllHooks` / `allowManagedHooksOnly`, a settings hot-reload regression on symlinked files, `claude --bg` connection drops at idle-exit, remote managed settings not retrying on 401, the managed `extraKnownMarketplaces` auto-update policy not being persisted, `Read` tool validation on whitespace-padded `offset` strings, and a Windows event-loop stall caused by missing executables.

Adds the agent view (Research Preview) for tracking every Claude Code session in one place, and a `/goal` command that keeps Claude working across turns until a completion condition is met. Hooks gain an exec form (`args: string[]`) and a `PostToolUse` `continueOnBlock` option, MCP stdio servers now receive `CLAUDE_PROJECT_DIR`, and `/mcp` Reconnect picks up `.mcp.json` edits without a restart. Note: Remote Control, `/schedule`, claude.ai MCP connectors, and notification preferences are now disabled whenever `ANTHROPIC_API_KEY` / `apiKeyHelper` / `ANTHROPIC_AUTH_TOKEN` is set, even alongside a Claude.ai login.

Agent view is a centralized interface inside the Claude Code CLI for managing multiple concurrent agent sessions. Now in Research Preview, it lets developers see which agents need input, are actively working, or have completed tasks — and scale up background work via peek, `/bg`, and `claude --bg`.

The Claude Platform on AWS — Anthropic-operated, accessed via AWS IAM with CloudTrail audit logging and a single AWS invoice — is generally available. AWS customers get day-one access to all native Claude API features, including Claude Managed Agents (beta), Advisor strategy (beta), Files API (beta), Skills (beta), and the MCP connector (beta), with Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 available at launch.

Session quality survey is re-enabled for OTEL enterprises via `CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL`, and `settings.autoMode.hard_deny` lets classifier rules block actions unconditionally. WSL2 image paste now falls back to PowerShell when xclip/wl-paste are unavailable. Plus a wide round of fixes — MCP servers disappearing after `/clear`, login loops from concurrent OAuth token writes, and plan mode failing to block writes when an `Edit(...)` allow rule matched.

A new `worktree.baseRef` setting (`fresh` | `head`) lets you choose whether worktrees branch from `origin/<default>` or local `HEAD`, and hooks and Bash commands can now read the active effort level via `$CLAUDE_EFFORT`. Plus fixes for a refresh-token race that dead-ended parallel sessions at 401, MCP OAuth ignoring `HTTP(S)_PROXY`/mTLS across discovery and token refresh, and subagents failing to discover project, user, or plugin skills.

`CLAUDE_CODE_SESSION_ID` is now exposed to Bash subprocesses, and `CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN=1` lets you opt out of the fullscreen renderer. Plus a wide round of stability fixes — graceful shutdown on external SIGINT, blank fullscreen after sleep/wake, 10GB+ RSS growth from stdio MCP servers, scroll regressions in Cursor/VS Code/JetBrains terminals, and Bedrock/Vertex 400s when `ENABLE_PROMPT_CACHING_1H` is set.

`--plugin-url` loads a `.zip` plugin straight from a URL, and Homebrew/WinGet installs can self-upgrade in the background. Ctrl+R history is back to searching all projects by default, and Gateway `/v1/models` discovery is now opt-in via env var. Plus fixes for 1-hour prompt cache TTL silently downgrading, OAuth logouts after wake-from-sleep, and `/context` wasting ~1.6k tokens per call.

A month after the April 8 beta launch, Managed Agents added two new features and promoted three from research preview to public beta. Dreaming (research preview) curates memory between sessions for self-improvement, and outcomes, multiagent orchestration, memory, and webhooks all ship as public beta.

Ten ready-to-run financial services agents — pitch builder, KYC screener, month-end closer, and more — ship as plugins in Claude Cowork and Claude Code, and as cookbooks for Claude Managed Agents. A Moody's MCP app brings proprietary credit ratings and data on 600M+ companies, and 17 data partners are integrated alongside.

The Claude add-ins for Excel, PowerPoint, and Word are generally available on all paid plans, and Claude for Outlook joins in public beta. Once installed, context carries automatically between apps so a model built in Excel can become a PowerPoint deck or a Word memo without re-explaining inputs, assumptions, or calculation flow.

`--channels` now works with console (API key) authentication, and `/mcp` shows per-server tool counts and flags zero-tool servers. `OTEL_*` environment variables no longer leak into subprocesses, and `EnterWorktree` branches from local HEAD as documented. Plus fixes for falsely-blocked 1M-context sessions, parallel shell tool regressions, sub-agent cache misses, and more.

Gateway-driven `/model` picker, `claude project purge` for full project state cleanup, OAuth code paste for WSL/SSH/container logins, broader `--dangerously-skip-permissions` scope, plus Windows PowerShell 7 detection, CJK rendering fix, and a long list of OAuth, stream-timeout, and Remote Control fixes.

Anthropic launched Claude Security in public beta — a Claude Opus 4.7-powered tool that scans codebases for vulnerabilities and generates targeted patches. Available directly at claude.ai/security and embedded in CrowdStrike, Microsoft Security, Wiz, and other platforms. Currently for Claude Enterprise customers; Claude Team and Max coming soon.

`ANTHROPIC_BEDROCK_SERVICE_TIER` to choose Bedrock service tier, `/resume` accepts PR URLs across GitHub/GHE/GitLab/Bitbucket, `/mcp` surfaces connectors hidden by duplicate URLs, plus fixes for `/branch` rewound-timeline forks, structured-output 400s on Vertex/Bedrock, and image resizing on newer models.

MCP `alwaysLoad`, `claude plugin prune`, `/skills` filter search, `PostToolUse` output replacement for all tools, scrollable overflow dialogs, and a large cleanup of memory leaks, scrollback duplication, and resume crashes.

AWS announced Claude Platform on AWS — a new way to use Anthropic's native Claude Platform with AWS credentials, billing, and access controls. It sits alongside Claude on Amazon Bedrock as a second path with different data-residency tradeoffs.

`/config` changes now persist to `~/.claude/settings.json`, `--from-pr` supports GitLab/Bitbucket/GitHub Enterprise, custom `prUrlTemplate`, PowerShell auto-approval parity with Bash, and a large cleanup of bugs and one security fix.

Vim visual and visual-line modes, custom themes, MCP-tool hooks, `/cost` and `/stats` merged into `/usage`, WSL inheriting Windows managed settings, and a long list of MCP OAuth fixes.

Opus 4.7 `/context` now uses the native 1M window, `/model` selections persist across restarts, `/resume` offers to summarize stale large sessions, native builds swap Glob/Grep for bfs/ugrep via Bash, and default effort rises to `high` for Pro/Max.

AWS published the official guide to running Claude Cowork on Amazon Bedrock — reuse the same Bedrock infrastructure you built for Claude Code to bring Claude to every knowledge worker, inside your AWS perimeter.

Up to 67% faster `/resume` on large sessions, quicker MCP startup, inline thinking progress indicator, smoother full-screen terminal scrolling, and a sandbox dangerous-path safety fix.

A Microsoft Word add-in that brings Claude into your document — with native tracked changes, template-aware drafting, and consistency checks.

Collaborative visual creation with Claude — designs, interactive prototypes, presentations, and more through conversation.

Native-binary distribution, network sandbox denied domains, full-screen viewport scrolling, parallelized `/ultrareview`, and security/stability fixes.

Opus 4.7 `xhigh` effort, `/ultrareview` cloud parallel review, `/effort` slider, Auto theme, and reduced permission prompts.

A new Opus model with significant gains in coding, high-resolution vision, and instruction following. Claude Code adds /ultrareview and the xhigh effort level.

Mid-conversation `/tui` mode switch, mobile push notifications, `/focus` command, and MCP connection stability fixes.

Autonomous routines in Claude Code triggered by schedules, API calls, and GitHub events. Run code review, bug fixes, and deploy verification from the cloud — laptop closed.

/recap for session summaries on return, built-in slash command auto-discovery, 1-hour prompt cache TTL option, plus stability fixes.

Plugin background monitors, PreCompact hook, automatic API stream-stall recovery, and improvements to /doctor.

/team-onboarding, OS cert store auto-trust, memory-leak fix, command-injection patch, and 40+ other fixes.

Vertex AI setup wizard, Perforce mode, Monitor tool, subprocess sandboxing, and significant Bash permission hardening.

Focus view toggle, status line refresh interval, improved CJK input, and a batch of stability fixes.

A new API alongside Messages that provides a pre-built agent harness and managed infrastructure — run Claude agents without building your own loop, sandbox, or tool-execution layer.

Keep a local Claude Code session running and control it from claude.ai/code or the mobile apps — with your local files, MCP servers, and tools intact.