Claude Report — 2026-06-29

🚀 Anthropic Official

US government clears Claude Mythos 5 for ~100 institutions (June 26-27)

The US Commerce Department lifted its export block on Claude Mythos 5, granting access to roughly 100 US companies and federal agencies after a two-week standoff driven by cybersecurity concerns; the letter from Commerce Secretary Howard Lutnick remains silent on Fable 5, and European access stays restricted, establishing a new government-oversight framework for frontier-model releases. Full exclusive: Semafor (June 27).

API rate limits raised across all Claude models (June 26)

Claude Sonnet and Claude Haiku rate limits now match Claude Opus at every usage tier; usage tiers are consolidated from the previous structure into three — Start, Build, and Scale — with no organization receiving lower limits than before and no action required; limits are visible in the Claude Console.

Claude Code v2.1.195 (June 26)

Adds CLAUDE_CODE_DISABLE_MOUSE_CLICKS env var to disable click/drag/hover in fullscreen mode while keeping scroll; fixes hook matchers with hyphenated identifiers (e.g. mcp__brave-search) that were substring-matching — they now exact-match, with mcp__brave-search__.* for wildcard; fixes voice dictation capturing silence on macOS after a device change and auto-submit never firing for space-free scripts (Japanese, Chinese, Thai); fixes external plugins bypassing install consent on some loader paths, /plugin Enable/Disable when plugin.json name differs from the marketplace entry name, background jobs vanishing from claude agents when written by a newer version, crashed tasks showing a blank screen for up to 5 seconds, and daemon sockets failing to start; improves Linux voice mode to distinguish "no mic" from "SoX missing"; improves claude agents completed list to fill vertical space; adds a provisioning checklist during Remote session container startup.

🔌 Claude Code Plugins

Plugin tracking skipped — ranking cache unavailable this cycle.

🛠️ Skills

None of the top 5 updated in window.

Unchanged in window: anthropics/skills, alirezarezvani/claude-skills, Piebald-AI/claude-code-system-prompts, glebis/claude-skills, ComposioHQ/awesome-claude-skills

🤖 Agents & Subagents

claude-agent-sdk-typescript v0.3.195 (June 26)

Adds Query.reinitialize() to re-send the initialize control request and redeliver pending permission/dialog prompts after a transport gap; fixes commands_changed event not firing for synced skills when the skill list resolves before the change-detector subscribes.

Unchanged in window: nicobailon/pi-subagents, Chachamaru127/claude-code-harness

🔗 MCPs & Integrations

github-mcp-server v1.5.0 (June 27)

Bakes OAuth 2.1 login into the STDIO server transport — personal access tokens are no longer required; adds reactions to issue and pull request comments; exposes parent-issue retrieval via the issue_read tool; and surfaces labels, milestones, and additional metadata in the MCP App for Issues.

mcp Python SDK v2.0.0a3 (June 26)

First alpha targeting the upcoming 2026-07-28 spec: stateless protocol is now negotiable end-to-end, protocol types are split into a standalone mcp-types package, multi-round tool calls land via InputRequiredResult, ServerMiddleware is reshaped to a (ctx, call_next) pattern, and OAuth client conformance is updated for RFC 9207 / SEP-837 / SEP-2350 / SEP-2352; roots, sampling, and logging methods deprecated per SEP-2577.

Unchanged in window: awslabs/mcp, modelcontextprotocol/typescript-sdk (main), portainer/portainer-mcp

💡 Community — Workflows & Ideas

Using Opus 4.8 to get a second opinion on an MRI (June 28)

Publishes a multi-agent DICOM analysis workflow in Claude Code with Opus 4.8 (xhigh): two independent subagents process the same shoulder MRI separately to avoid cross-contamination, surfacing a diagnostic discrepancy — clinic said grade III partial tear; Claude suggested mild insertional tendinosis with no discrete tear — and sparking 575 HN comments (437 pts) on AI reliability in medical imaging.

📰 Quick Mentions

Relevant links that don't warrant a section of their own:

• modelcontextprotocol/servers security patch (June 29) — bumps python-multipart, cryptography, and pyjwt to clear HIGH-severity dependency alerts
• Ask HN: Anthropic banned me from using Claude Code (June 26) — active thread on account suspension edge cases and appeal paths; no policy change announced