Claude docs changes for June 13th, 2026 [diff]

Executive Summary

• Subagents can now spawn their own subagents (as of v2.1.172), enabling multi-level agent delegation with depth limits for background agents
• New Analytics APIs guide clarifies which of the two analytics APIs (Claude Code vs. Claude Enterprise) fits your organization and how to provision keys
• AWS region configuration for Bedrock is now optional if your AWS profile already sets a region, with a full resolution order including AWS config files
• The availableModels allowlist now covers subagents and the advisor tool (not just the main session), and matching now supports version prefixes and full model IDs
• New CLAUDE_CODE_CHILD_SESSION environment variable lets scripts reliably detect when they're running inside a Claude Code tool call or hook

New Claude Code versions

2.1.176

New features

• Session titles are now generated in the language of your conversation; set the language setting to pin a specific language
• Added footerLinksRegexes setting for regex-matched link badges in the footer row, configurable via user or managed settings

Existing feature improvements

• Improved Bedrock credential caching: credentials from awsCredentialExport are now cached until their actual Expiration timestamp instead of a fixed 1 hour
• Background sessions now show clearer guidance when a window left open across an auto-update can't submit a reply, and claude daemon status explains version-skew behavior

Major bug fixes

• Fixed availableModels enforcement: alias model picks can no longer be redirected to a blocked model via ANTHROPIC_DEFAULT_*_MODEL environment variables, and /fast now refuses to toggle when it would switch to a blocked model
• Fixed auto mode failing on Fable 5 for organizations without Opus 4.8 enabled — the classifier now falls back to the best available Opus model
• Fixed hook if conditions for Read/Edit/Write tool paths: patterns like Edit(src/**), Read(~/.ssh/**), and Read(.env) now match correctly
• Fixed Linux sandbox failing to start when .claude/settings.json is a symlink with an absolute target
• Fixed /copy and mouse-selection copy not reaching the system clipboard inside tmux over SSH, and tmux paste buffer not loading on versions older than 3.2
• Fixed Remote Control connecting from web/mobile silently switching the session's model
• Fixed Remote Control disconnect notifications showing a bare numeric code instead of a human-readable reason, and connection failures adding a duplicate line to the transcript
• Fixed Remote Control sessions not disconnecting when you sign in to a different account
• Fixed /cd and worktree moves leaving the session reporting the previous directory's git branch
• Fixed claude agents: pressing back in one window no longer detaches other windows attached to the same session
• Fixed backgrounded sessions showing "Working" forever when /bg mid-turn had nothing left to continue
• Fixed background agent search by PR URL: PRs opened during scheduled wakeups or while a job was blocked now appear in claude agents search
• Fixed claude --bg -cn <name> not seeding the session name
• Fixed cloud sessions failing with "Could not resolve authentication method" when idle for too long before being claimed

Claude Code changes

Changed documents

Agent Loop [Source]

• Updated xhigh effort level recommendation from "Opus 4.8 and Opus 4.7" to "Fable 5 and Opus 4.7+" [line 136] [Source]
• Expanded error_max_structured_output_retries description to cover model fallback retracting completed output as a second failure mode [line 237] [Source]

Structured Outputs [Source]

• Added explanation that model fallbacks can retract completed structured output mid-stream, causing the same error as validation failures; advise checking errors text to distinguish the two causes [line 216] [Source]

Subagents (Agent SDK) [Source]

• As of v2.1.172, subagents can spawn their own subagents; foreground subagents can spawn at any depth, background subagents are limited to five levels deep; updated note and added pointer to full depth rules [line 117] [Source]

TypeScript SDK Reference [Source]

• Added skipMcpDiscovery optional field to SdkPluginConfig; when true, the SDK loads plugin skills/hooks/agents/commands but skips reading its MCP server configuration [line 64] [Source]
• Added auto-continuation origin kind to SDKMessageOrigin for synthetic turns injected when a session continues without fresh user input [line 82] [Source]
• Added resolvedModel optional field on AgentOutput completed and async_launched variants, naming the model the subagent actually ran on; requires v2.1.174 [line 2101] [Source]

Amazon Bedrock [Source]

• AWS_REGION is now optional as of v2.1.172; Claude Code resolves the region from AWS_REGION, AWS_DEFAULT_REGION, the active AWS profile, then us-east-1 as a fallback; older versions still require AWS_REGION explicitly [line 148] [Source]
• Added note that AWS GovCloud regions should use the us-gov. prefix in model IDs [line 165] [Source]

Costs [Source]

• Added note that the VS Code extension shows the same usage breakdown as /usage via the Account & usage dialog with Day/Week toggle; requires v2.1.174 [line 22] [Source]

Environment Variables [Source]

• Updated CLAUDECODE description: IDE extensions also set this in their integrated terminals; added guidance to use CLAUDE_CODE_CHILD_SESSION instead when you need to distinguish nested sessions [line 211] [Source]
• Added new CLAUDE_CODE_CHILD_SESSION variable (v2.1.172): set to 1 in Bash/PowerShell tool calls, hook commands, and status line commands — not set in stdio MCP servers or IDE terminals, making it a reliable way to detect a direct tool subprocess [line 225] [Source]
• Updated CLAUDE_AUTOCOMPACT_PCT_OVERRIDE with detailed explanation of when it takes effect (cloud sessions, Remote Control, Sonnet/Opus 4.6 models, and when CLAUDE_CODE_AUTO_COMPACT_WINDOW is set) [line 217] [Source]
• Updated CLAUDE_CODE_FORCE_SESSION_PERSISTENCE — reinstated to override CLAUDE_CODE_CHILD_SESSION misclassification; clarified behavior across version ranges [line 234] [Source]
• Updated CLAUDE_CODE_SCROLL_SPEED to accept fractional values below 1 (e.g., 0.5) to slow accelerated trackpad scrolling [line 243] [Source]

Errors [Source]

• Added "Could not resolve authentication method" to the error navigation table and added a new section explaining it — affects background, cloud, and SDK sessions; upgrade to v2.1.174 to fix the idle worker variant [line 255] [Source]
• Added note for the 1M context entitlement error: Claude Code now auto-compacts when context exceeds 200K and keeps the session at that limit; v2.1.171 and earlier required /clear to recover [line 263] [Source]

Fullscreen Rendering [Source]

• Updated scroll speed documentation: CLAUDE_CODE_SCROLL_SPEED now accepts fractional values below 1 to slow trackpad acceleration [line 308] [Source]
• Added note about wheelScrollAccelerationEnabled setting (v2.1.174): set to false for a constant scroll rate per wheel notch [line 310] [Source]

Hooks [Source]

• Clarified that model field in SessionStart hooks is not guaranteed to be present (e.g., after /clear or conversation recovery) — check for the field before reading it [line 322] [Source]
• Added resolvedModel field to PostToolUse Agent tool response (v2.1.174): names the actual model the subagent ran on, present on both completed and async_launched status variants [line 340] [Source]

JetBrains [Source]

• Rewrote installation instructions to clearly explain that the plugin does not bundle the CLI — both the CLI and the plugin must be installed separately; added note about setting the claude path if not on PATH [line 23] [Source]

Model Configuration [Source]

• Expanded availableModels enforcement: the allowlist now applies to the main session, subagent models, the advisor model, and fallback chains [line 90] [Source]
• Updated opusplan context window behavior: it now receives the same 1M context upgrade as opus on subscription tiers that auto-upgrade; the opusplan[1m] suffix works to force 1M for both plan and execute phases [line 166] [Source]
• Updated availableModels matching: now matches on alias, version prefix (e.g., claude-opus-4-8), or full model ID; [1m] suffix is stripped before matching [line 469] [Source]

Monitoring Usage [Source]

• Added model attribute to the claude_code.lines_of_code.count metric (v2.1.172), enabling per-model breakdown of code additions/removals [line 482] [Source]
• Added host_owned_mcp field to plugin_loaded event (v2.1.172): true when the SDK host manages the plugin's MCP connections [line 490] [Source]

Permissions [Source]

• Clarified glob pattern behavior: * matches within a single path segment and can appear at any position, while ** matches across directories [line 534] [Source]
• Added comprehensive WebFetch domain matching rules: * as a leading *. matches subdomains at any depth; WebFetch(domain:*) matches every domain; exact rules take precedence over wildcards [line 543] [Source]

Plugin Hints [Source]

• Added CLAUDE_CODE_CHILD_SESSION (v2.1.172) as an alternative to CLAUDECODE for gating plugin hint emission; explains trade-offs: CLAUDECODE has broader reach, CLAUDE_CODE_CHILD_SESSION is less likely to appear in human terminals [line 562] [Source]

Remote Control [Source]

• As of v2.1.172, the Remote Control footer indicator reads /rc active (shortened) and is hidden when the terminal is too narrow; earlier versions always showed Remote Control active [line 61] [Source]

Settings [Source]

• Updated availableModels description to mention subagents and the advisor tool (not just the main session) [line 621] [Source]
• Added wheelScrollAccelerationEnabled setting (v2.1.174): controls whether scroll speed accelerates during fast scrolling; default true [line 629] [Source]

Sub-agents [Source]

• Added new "Spawn nested subagents" section (v2.1.172): subagents can now spawn their own subagents; foreground subagents can spawn at any depth, background subagents are capped at depth 5; subagent panel shows the full tree with descendant counts [line 678] [Source]
• Removed Agent from the list of tools unavailable to subagents — subagents can now use the Agent tool to spawn children [line 285] [Source]
• Updated /agents Running tab description: now lists both live and recently finished subagents [line 145] [Source]

VS Code [Source]

• Added new "Check account and usage" section: /usage opens an Account & usage dialog with plan, usage bars, and attribution tables for skills/subagents/plugins/MCP servers; includes Day/Week toggle; requires v2.1.174 [line 118] [Source]

API changes

New Documents

Analytics APIs [Source]

New guide explaining the two analytics APIs: the Claude Code Analytics API (uses an Admin API key, reports Claude Code productivity metrics for Claude Platform organizations) and the Claude Enterprise Analytics API (uses an Analytics API key created in claude.ai, reports organization-wide engagement, adoption, and cost data for Claude Enterprise organizations). Includes a comparison table, step-by-step instructions for provisioning each key type, and a summary of what each API covers.

Changed documents

C# Beta: File Upload [Source]

• Page was previously a "Console temporarily unavailable" placeholder; now contains full C# SDK reference for the Beta.Files.Upload method including parameters, return type, and example code

C# Beta: Delete a Memory [Source]

• Page was previously a "Console temporarily unavailable" placeholder; now contains full C# SDK reference for Beta.MemoryStores.Memories.Delete including parameters, return type, and example code

Claude Code Analytics API [Source]

• Added callout noting that Claude Enterprise organizations should use the Claude Enterprise Analytics API (with an Analytics API key) for claude.ai users; links to the new Analytics APIs guide [line 26] [Source]

Compliance API Access [Source]

• Updated Analytics API key row in the key types table to link to the new Analytics APIs guide instead of the external support article [line 18] [Source]

Self-Hosted Sandboxes [Source]

• Added GKE Agent Sandbox to the list of platform-specific guides for managed agent sandboxes [line 31] [Source]

Usage and Cost API [Source]

• Added "Which API do you need?" section with a table distinguishing Claude Console (Admin API key) vs. Claude Enterprise (Analytics API key) for usage and cost reporting; links to the new Analytics APIs guide [line 20] [Source]

WIF Providers: Azure [Source]

• Clarified that aud claim behavior differs between Azure Entra v1 and v2 tokens: v2 tokens carry the application client ID (GUID), while v1 tokens carry the resource identifier passed when fetching the token [line 56] [Source]