TLDR;
Anthropic shipped Claude Tag — a multiplayer @Claude that lives in your Slack channels for Enterprise and Team customers — and published a new Economic Index report on how people actually pace their Claude usage. Claude Code had a fast week: releases v2.1.186 through v2.1.195 hardened auto-mode safety, MCP authentication, voice dictation, and background agents (plus a quiet OpenTelemetry change that can start logging assistant responses on upgrade — worth a look before you update). On the competitive side, OpenAI's Codex Remote reached general availability and OpenAI previewed the GPT-5.6 (Sol / Terra / Luna) family. One deprecation to flag: claude-mythos-preview retires Tuesday, June 30 — its successor claude-mythos-5 stays limited to Project Glasswing partners, so general users should build on Claude Opus 4.8.
Sponsored by CodeRabbit
Review the actual change, not the file list
AI writes more code than ever. Reviewing it shouldn't mean scrolling forty files in alphabetical order.
CodeRabbit Review reorganizes any pull request from a flat file list into a structured, layer-by-layer walkthrough — the logical reading order of the change, not the order your platform happens to sort it. Every range gets a plain-language summary, with sequence diagrams, state machines, and ERDs generated inline wherever a visual earns its place.
Cohorts group related files so you review one idea at a time. Layers put foundational changes — data shapes, contracts — before the code that depends on them. Code Peek shows any symbol's definition and usages without leaving the tab, and Semantic Diff cuts past formatting noise to what actually changed. Comment on exact line ranges; native reviews and approvals post back to GitHub or GitLab.
From the team that pioneered AI code reviews: 2M reviews a week, 6M repos, 15K customers. Free for everyone in early access.
Review your next PR with CodeRabbit Review →
API & Model Updates
Jun 23, 2026
Claude Tag lets Claude Enterprise and Team customers add @Claude to selected Slack channels, grant it scoped access to tools, data, and even codebases, and then tag it to delegate tasks. One Claude per channel means anyone can see what it's working on and pick up where a teammate left off; it builds context from the channels it's in over time and can schedule work to run later. It replaces the old "Claude in Slack" app (admins opt in to migrate within 30 days) and runs on Opus 4.8. Anthropic frames it as the next step beyond Claude Code — 65% of its own product team's code now comes from an internal version. If you're integrating Claude into team workflows, the multi-user session model and channel-scoped permissions are the parts to study.
Jun 23, 2026
Org admins can now restrict which models appear in the model picker, the --model flag, the /model command, and the ANTHROPIC_MODEL env var — users hitting a restricted model see a clear "restricted by your organization's settings" message. If your deployment uses managed settings, audit your availableModels policy (enforced via enforceAvailableModels) before the next auto-update.
Claude Code & CLI
A run of releases landed this week, v2.1.186 through v2.1.195 (several version numbers were skipped). Highlights by theme:
Jun 25, 2026
Auto-mode denial reasons now appear in the transcript, the denial toast, and /permissions recent denials — no more silent rejections. The new autoMode.classifyAllShell setting routes all Bash/PowerShell commands through the classifier, not just arbitrary-code-execution patterns; useful for tighter enforcement in shared or CI environments.
Jun 25, 2026
v2.1.193 added a claude_code.assistant_response OpenTelemetry log event that carries the model's response text. It's redacted by default — but when OTEL_LOG_ASSISTANT_RESPONSES is unset, it follows your existing OTEL_LOG_USER_PROMPTS setting. So any deployment already logging user prompts will silently start capturing assistant response content on upgrade. Set OTEL_LOG_ASSISTANT_RESPONSES=0 to keep prompts-only logging.
Jun 24, 2026
/rewind can now resume a conversation from before /clear was run, giving you a durable escape hatch when you need to recover context you thought was gone. The same release permanently fixes background agents resurrecting after being stopped from the tasks panel, cuts CPU use during streaming by roughly 37% (text updates are coalesced to 100ms), and fixes comma-separated hook matchers (e.g. "Bash,PowerShell") that previously never fired.
Jun 22–26, 2026
claude mcp login and claude mcp logout now authenticate MCP servers without opening the /mcp menu, with --no-browser stdin redirect to complete cleanly over SSH (v2.1.186). Hook matchers with hyphenated identifiers (e.g. mcp__brave-search) now exact-match instead of substring-match — use mcp__brave-search__.* to match all tools from that server (v2.1.195; pairs with the v2.1.191 comma-matcher fix). And structured output via --json-schema and agent({schema}) no longer loops indefinitely after a successful call (v2.1.187).
Jun 22–25, 2026
! bash commands now get live file-path autocomplete (v2.1.193), and Claude automatically responds to the output of ! commands (v2.1.186) — set "respondToBashCommands": false in settings.json to keep the previous context-only behavior.
Jun 26, 2026
Voice dictation on macOS no longer captures silence during long sessions when the default input device changes. Auto-submit now fires correctly for languages written without spaces (Japanese, Chinese, Thai). Linux voice mode now distinguishes "no microphone" from "SoX not installed."
Jun 23–25, 2026
The new sandbox.credentials setting blocks sandboxed commands from reading credential files and secret env vars — useful defence-in-depth in multi-agent pipelines (v2.1.187). Idle background shell commands now get automatically reaped under memory pressure (disable with CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1) (v2.1.193). Background subagents also now surface permission prompts in the main session instead of auto-denying (v2.1.186).
Jun 23, 2026
Remote MCP tool calls that produce no response now abort with an error after 5 minutes instead of blocking indefinitely. Override the timeout with CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT if your server legitimately needs longer.
Jun 26, 2026
Disables mouse click/drag/hover in fullscreen mode while preserving wheel scroll — useful for terminal multiplexers or remote sessions where mouse events interfere with the host. (Context: v2.1.187 added mouse-click support to fullscreen select menus like permission prompts, /model, and /config; this env var turns it back off.)
Agent SDK & Managed Agents
Jun 26, 2026
Query.reinitialize() re-sends the initialize control request and redelivers pending permission/dialog prompts after a transport gap — helpful for long-running or resumable agent sessions that drop and reconnect.
Jun 25, 2026
The new promptSuggestions option on the Browser SDK's query() opts the remote CLI into emitting follow-up suggestions; this release also fixes console window flashes on Windows when spawning CLI subprocesses.
Jun 24, 2026
SDKRateLimitInfo.rateLimitType now includes seven_day_overage_included, and usage responses expose a model_scoped array with per-model weekly limit windows, utilization percentages, and reset times. Also fixes fast mode reverting to standard after the first turn.
Jun 22, 2026
Background agents now forward permission prompts to canUseTool instead of auto-denying, and stdin stays open while background tasks run. The rewind_conversation control request adds durable resume-anchor support for conversation rewinds. ReadMcpResourceDirTool is now a dedicated tool type.
Jun 24, 2026
pip install claude-agent-sdk==0.2.110 bundles Claude Code CLI v2.1.191.
MCP Ecosystem
Jun 25, 2026
Claude Code now shows a startup notice when any configured MCP server requires authentication and points users directly to /mcp — reducing silent failures in fresh environments.
Jun 22, 2026
MCP resource directory listing is now its own tool type in SDK schemas rather than a fallback inside ReadMcpResourceTool — relevant if your SDK code pattern-matches on tool types for logging or permission filtering.
Developer Tools & Community
Jun 26, 2026
This community repository mirrors Claude Code's system prompts within minutes of each release, and now tracks v2.1.195. It holds 515 prompt strings in total — builtin tool descriptions, sub-agent prompts, and utility prompts — with a CHANGELOG.md spanning 221 versions back to v2.0.14. Useful for anyone building tools that depend on understanding Claude Code's internal prompt structure.
Competitor Dev Tools
Jun 25, 2026
Codex Remote — start or continue work on a connected Mac or Windows host from the ChatGPT mobile app, reviewing progress and approving actions from your phone — is now generally available. Remote Control uses authenticated one-to-one QR pairing between each iOS/Android device and each host; connections used since June 8 stay paired, while older inactive ones need to re-pair. The same update ships a DigitalOcean plugin to provision a Droplet as a remote workspace. It closes the "drive your coding agent from your phone" gap that Claude Code doesn't cover natively — and the forced re-pairing of pre-June-8 connections is worth flagging to anyone who left old pairings live.
Jun 26, 2026
OpenAI previewed the GPT-5.6 family: Sol (the flagship, its strongest model yet), Terra (balanced everyday work at roughly GPT-5.5 quality for about half the price), and Luna (fast and cheapest). The naming now separates generation (the number) from durable capability tiers (Sol/Terra/Luna). A new max reasoning effort and a new ultra mode that fans work out to subagents target complex agentic work, and Sol sets state-of-the-art on Terminal-Bench 2.1. Pricing per 1M tokens: Sol $5 in / $30 out, Terra $2.50 / $15, Luna $1 / $6. Important caveat: this is a limited preview — access is gated to a small set of trusted partners via the API and Codex (a US-government-requested phased rollout tied to cyber capability), with general availability "in the coming weeks." For most builders it's watch-don't-deploy; the nearer-term pressure is Terra's price, not Sol's availability.
Research & Safety
Jun 26, 2026
The latest Economic Index report examines usage cadences — how frequently, at what times of day, and in what weekday/weekend patterns people use Claude for work — drawing on usage data plus a roughly 9,700-person survey. Useful background for builders designing products around sustained Claude usage rather than one-shot interactions.
Action Items
By Tuesday, June 30, 2026:
-
claude-mythos-preview retires Tuesday, June 30. Anthropic's docs list claude-mythos-5 as the successor, but it's restricted to vetted Project Glasswing partners — and Mythos/Fable access has been export-suspended since June 12. For general production use, build on Claude Opus 4.8 (claude-opus-4-8); Claude Fable 5 (claude-fable-5) is the same-tier public option once access is restored.
Ongoing — Review Before Next Deploy:
-
Check OpenTelemetry logging before upgrading to v2.1.193+ — if you log user prompts but don't want assistant response text captured, set
OTEL_LOG_ASSISTANT_RESPONSES=0 (otherwise it inherits your prompt-logging setting on upgrade).
-
Update MCP hook matchers with hyphenated server names — after v2.1.195,
mcp__brave-search exact-matches only; use mcp__brave-search__.* to catch all tools from that server.
-
Audit your
availableModels managed-settings policy — v2.1.187 enforces org model restrictions across the picker, --model, /model, and ANTHROPIC_MODEL.
-
Review the
sandbox.credentials setting — new in v2.1.187; enable it in multi-agent or shared environments to block sandboxed commands from reading credential files and secret env vars.
-
Set
"respondToBashCommands": false if you rely on ! bash commands being context-only — v2.1.186 now triggers Claude to respond to ! output automatically by default.
-
Check
CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT if any remote MCP server legitimately needs more than 5 minutes to respond — the new v2.1.187 default will abort those calls with an error.
-
Upgrade the TypeScript Agent SDK to v0.3.191+ to get per-model weekly limit data (
model_scoped) in usage responses — important for dashboards or quota-management tooling.
By August 5, 2026:
-
Migrate off
claude-opus-4-1-20250805 — deprecated June 5 with a 60-day clock; hard retirement August 5; migrate to Claude Opus 4.8.
Want to catch up? Read every past issue of This week in Claude.
Did you like this issue?
One tap is the fastest way to shape next week.
👍 Liked it · 👎 Didn't like it
And I read every reply. Hit reply and tell me what you want more of, or less of, and it goes straight into the next issue.
