Claude docs changes for July 22nd, 2026 [diff]

Executive Summary

• Claude Code Desktop adds an iOS Simulator pane (public beta, macOS): Claude can build, install, launch, and tap through iOS apps while you watch or take over yourself.
• /verify and /code-review no longer run automatically — Claude now only invokes them when you explicitly ask, keeping longer-running checks under your control.
• A new sandbox.filesystem.disabled setting lets you keep network isolation while turning off filesystem isolation entirely, for workloads that need broad file access but confined egress.
• 2.1.217 adds emoji shortcode autocomplete, a cap on concurrently-running subagents (default 20), and stops subagents from spawning nested subagents by default.
• The Claude API's search-results feature (RAG citations) now works on all active models with no beta header required, and the docs clarify several validation rules (text-only content, user-message-only, tool-result homogeneity).

New Claude Code versions

2.1.208

Existing feature improvements

• Changed the Fable 5 usage-credits consent prompt to start with the decline option focused

2.1.217

New features

• Added emoji shortcode autocomplete in the prompt input: type :heart: to insert ❤️, or :hea for suggestions — disable with the emojiCompletionEnabled setting
• Added warnings when transcript writes are failing (e.g. disk full) or when session saving is off due to an inherited environment variable, instead of losing transcripts silently
• Added a cap on concurrently-running subagents (default 20, override with CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS) so one message can't fan out unbounded background agents
• Changed subagents to no longer spawn nested subagents by default; set CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH to allow deeper nesting

Existing feature improvements

• Improved footer PR badge links to be clickable hyperlinks even when terminal support can't be detected (e.g. over ssh/tmux); set FORCE_HYPERLINK=0 to opt out
• Changed the login-expiry warning to appear 3 days before expiry instead of 5
• Capped the frontend-design plugin suggestion tip at 3 lifetime impressions instead of repeating indefinitely

Major bug fixes

• Fixed a memory leak where truncated MCP tool outputs kept the full untruncated result in memory for the rest of the session
• Fixed Windows auto-update failures that could leave claude.exe missing; failed updates now restore the preserved executable automatically
• Fixed background session isolation not canonicalizing symlinked working directories, which could let sessions escape their workspace folder
• Fixed auto-compact never triggering for Claude Opus 4.8 on Bedrock and /compact failing once over the limit
• Fixed corporate mTLS, TLS-verify, OAuth scope, and proxy settings being ignored in Claude Desktop sessions
• Fixed screen reader mode's startup announcement being cut off by the first prompt render, and the thinking status row re-rendering every few seconds to update elapsed time and token counts
• Fixed managed settings that set OTEL_EXPORTER_OTLP_ENDPOINT not governing all signals — lower-scope signal-specific overrides no longer redirect telemetry away from the managed endpoint
• Fixed --resume/--continue and /resume failing with a TypeError when a transcript has a malformed attachment entry
• Fixed Remote Control sessions not showing a pending permission prompt or dialog to viewers that connected after it appeared
• Fixed background shells sometimes becoming impossible to stop after a session is sent to the background or when the session exits on a heavily loaded machine, most visible on Windows
• Fixed a CLAUDE.md or SKILL.md paths frontmatter value with many brace groups OOM-killing or stalling the CLI at startup
• Fixed the transcript preview sitting flush against the input area when attaching to a starting background session
• Fixed --max-budget-usd not stopping background subagents: once the cap is reached, new spawns are denied and running background agents are halted

Claude Code changes

New Documents

desktop-ios-simulator [Source]

Documents the new iOS Simulator pane in Claude Code Desktop (public beta on macOS, Pro/Max/Team plans). The pane opens automatically when Claude builds, installs, launches, or checks an app in the simulator, streaming the device live. Covers requirements (Xcode, a Mac), interacting with the device yourself (tap, rotate, screenshot/record), how sessions manage devices, granting Claude access, the disableMobileSimulatorTools managed setting, and troubleshooting.

Changed documents

agent-sdk/observability [Source]

• OTEL_LOG_TOOL_CONTENT and OTEL_LOG_RAW_API_BODIES truncation is now configurable via CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH (requires v2.1.214+), instead of a fixed 60 KB. [lines 224-225] [Source]

channels [Source]

• Plugin installs for channels now prompt to choose the user scope so the plugin is available across all projects. [line 37] [Source]
• --channels and --dangerously-load-development-channels don't appear in claude --help during the research preview, even though they work. [line 324] [Source]

checkpointing [Source]

• New section: checkpointing doesn't rewind symlinked or hard-linked files — /rewind now skips them and shows a Restored the code, but skipped N files warning instead of silently writing through the link. [line 79] [Source]

chrome [Source]

• Chrome integration now requires signing in with /login — API key or setup-token sessions keep it off instead of silently failing every connection attempt with a 403. [line 30] [Source]
• New section on the "Claude wants to use your browser" install prompt shown when Claude needs the browser but doesn't detect the extension. [line 64] [Source]

cli-reference [Source]

• --channels now also accepts a Console API key for authentication, not just claude.ai. [line 62] [Source]
• --settings now requires the file to be a regular file no larger than 2 MiB. [line 110] [Source]

commands [Source]

• /verify and /code-review now run only when explicitly invoked; before v2.1.215 Claude could also run them on its own. [line 23] [Source]
• /context now shows a warning with how far over the context window you are and which command frees space, once the conversation exceeds it (v2.1.216+). [line 50] [Source]

computer-use [Source]

• Clarifies that in the Desktop app, iOS Simulator tasks now open the dedicated Simulator pane instead of screen control; from the CLI, computer use is still how Claude reaches the simulator. [lines 26, 160]

costs [Source]

• The in-CLI usage-credits management dialog (buy credits, set spend limit, configure auto-reload) has been replaced: /usage-credits now just opens billing settings in the browser, based on your role (Pro/Max, Team/Enterprise with or without billing access). [lines 29-33] [Source]

desktop [Source]

• Documents the new iOS Simulator pane and the disableMobileSimulatorTools managed setting to block it. [lines 149, 614]
• New url field for preview configurations, letting a preview open an address other than http://localhost:<port> (local HTTPS, *.localhost subdomains, or attaching to a server you already run). [lines 408-429] [Source]

discover-plugins [Source]

• Notes that /reload-plugins is needed after installing a plugin to activate it in the current session. [line 268] [Source]

errors [Source]

• Consolidated the spend-limit error entries into one Could not update your spend limit section, since /usage-credits no longer has its own dialog for this. [lines 346-354] [Source]
• Rewind's skipped-files guidance now suggests find . -type l / find . -type f -links +1 to locate skipped symlinked/hard-linked paths. [line 1378] [Source]

headless [Source]

• Streaming JSON output now waits up to 30 seconds (scaled to queue size) for a slow consumer to drain before exiting, up from a ~2 second cap in earlier versions that could cut off large responses. [line 135] [Source]

hooks [Source] / hooks-guide

• Behavior change: a PreToolUse prompt/agent hook that denies a call now ends the turn and shows the reason as a warning by default, instead of feeding it back to Claude as a tool error. Set continueOnBlock: true to restore the old (pre-v2.1.210) behavior. [hooks.md#L2713, hooks-guide.md#L747]
• Clarifies TaskCompleted hook behavior differs depending on whether it fires from a task completion (tool-error semantics) or a teammate stopping (TeammateIdle-like halt). [hooks.md#L2717]

how-claude-code-works [Source]

• Softened "every file edit is reversible" to "file edits are reversible" and notes that restores skip symlinked/hard-linked files. [lines 120-121] [Source]

interactive-mode [Source]

• Fixed: double Esc at an empty prompt could stop opening the rewind menu for the rest of a long-running session that used background tasks. [line 30] [Source]
• Fixed vim-mode . (repeat last change): change operators like cw/cc now replay the deletion too, and . after p/P repeats the paste instead of the previous change. [line 182] [Source]

jetbrains [Source]

• /ide now confirms a successful connection with a message like Connected to IntelliJ IDEA., and auto-installs the plugin for a detected IDE that lacks it. [line 63] [Source]

mcp [Source]

• Fixed: a failed list_changed refresh now keeps the server's previously discovered tools/prompts/resources instead of replacing them with an empty list. [line 160] [Source]
• Tools requiring anthropic/requiresUserInteraction now also withhold one-tap approval for any permission request that only the terminal dialog can render in full (safety warnings, always-allow options). [line 970] [Source]

memory [Source]

• Memory files with frontmatter now get a modified ISO 8601 timestamp recorded on each write, showing how current a fact is. [line 347] [Source]
• /memory opening a file in a GUI editor (e.g. VS Code) no longer blocks the session until the file is closed. [line 356] [Source]

model-config [Source]

• Fallback model chains now also skip a model with a smaller context window than the primary during compaction, since summarizing there would cut off part of the conversation; if every fallback is smaller, compaction shows the original error. [line 322] [Source]

monitoring-usage [Source]

• New CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH env var configures the OTel content truncation limit (default 60 KB) for tool content, raw API bodies, and other content-bearing attributes. [line 82] [Source]
• Fixed: the Prometheus-only metrics scrape no longer includes OpenMetrics-only # UNIT lines that some scrapers rejected. [line 432] [Source]
• Fixed: streams carrying usage across multiple frames (behind a proxy) no longer inflate claude_code.cost.usage / claude_code.token.usage. [line 1018] [Source]
• New message.uuid, client_request_id, tool_source, and cost_usd_micros attributes added across several event types for finer-grained correlation with the session transcript.

plugin-marketplaces [Source]

• claude plugin marketplace list --json now includes an installLocation field with the local cache path. [line 948] [Source]
• Troubleshooting now recommends git ls-remote <marketplace-url> to test git authentication directly. [line 1042] [Source]

remote-control [Source]

• /usage-credits from Remote Control now prints the billing URL instead of opening the removed in-CLI dialog. [line 237] [Source]

sandboxing [Source]

• New sandbox.filesystem.disabled setting turns off filesystem isolation while keeping network isolation on — useful when you want to control where sandboxed commands connect without restricting what they write. Only settable from user/managed settings or --settings, not project settings. [line 167] [Source]

scheduled-tasks [Source]

• Scheduling a task now fails with an error if the project's .claude directory or the task file is a symlink, instead of writing through the link. [line 196] [Source]

sessions [Source]

• Resuming a session now also restores the agent it was started with (--agent or the agent setting), including its system prompt, tool restrictions, and model. [line 26] [Source]
• A session that fails to load from claude --resume now prints Failed to resume the conversation with a retry command and exits, instead of failing silently. [line 86] [Source]

skills [Source]

• /verify and /code-review now run only when explicitly invoked, not automatically. [line 13] [Source]
• Fixed: a plugin skill's frontmatter name now only replaces the last segment of the command (keeping the plugin prefix and adding a bare alias), instead of replacing the whole command name. [line 246] [Source]

API changes

New Documents

get-api-key [Source]

New page walking through creating and using a Claude API key: signing in to the Console, creating a key, storing it, setting ANTHROPIC_API_KEY, and how this relates to the Admin API's key-management endpoints.

Changed documents

api/rate-limits [Source]

• Clarifies that new organizations or those with limited usage history may start with limits below the standard tier limits shown on the page, increasing automatically as usage history builds. [line 22] [Source]

build-with-claude/search-results [Source]

• Search results with citations now work on all active models except Claude Haiku 3, with no beta header required — previously limited to a specific model list. [line 13] [Source]
• Clarifies source can be any stable string, not just a URL (e.g. kb://article-1234). [line 54] [Source]
• New validation rule: within a tool_result, if any block is a search_result, all blocks must be search_result — mixing types returns an error. [line 511] [Source]
• New limitation: search_result blocks can only appear in user messages (including inside tool results); assistant messages with them are rejected. [line 588] [Source]

cli-sdks-libraries/cli/scripting [Source]

• New section on authenticating curl requests with ant auth login CLI credentials instead of a static API key, using ant auth print-credentials --access-token. [line 239] [Source]

manage-claude/cmek [Source]

• Clarifies which roles can configure CMEK: Organization Admins on Claude Platform, or Owners/Primary Owner on Claude Enterprise. [line 23] [Source]
• Claude in Chrome added to the list of surfaces CMEK covers. [line 66] [Source]

manage-claude/cmek-aws-kms [Source]

• Clarifies that key validation always sends the all-zeros compartment UUID as the encryption context (since validation runs before the key is attached to a workspace), while live traffic sends the real compartment ID — the key policy's EncryptionContext condition must allow both. [line 140] [Source]