Claude Report — 2026-06-18

🚀 Anthropic Official

Claude Code v2.1.181 (June 17)

Adds /config key=value to set any setting from the prompt (works in interactive, -p, and Remote Control modes); introduces sandbox.allowAppleEvents for macOS Apple Events from sandboxed commands; adds CLAUDE_CLIENT_PRESENCE_FILE to suppress mobile push notifications while at the machine; upgrades bundled Bun to 1.4; streams long paragraphs line-by-line instead of buffering. Auto-retry now covers mid-thinking connection drops; the subagent panel auto-hides idle agents after 30 s with a 5-row cap and scroll hints; MCP OAuth browser page now matches Claude Code's visual style and auto-closes on success. Fixes 30+ issues including 0-byte writes on network drives, prompt-cache misses on custom ANTHROPIC_BASE_URL and Foundry, startup crashes with corrupted .claude.json, macOS TUI freeze during Spotlight reindex, long-session history loss during 30-day transcript cleanup, and unbounded foreground subagent nesting (depth now capped at 5 levels).

Claude Code v2.1.178 (June 15)

Introduces Tool(param:value) permission syntax — e.g., Agent(model:opus) blocks Opus subagent spawns and * wildcards match broadly. Skills in nested .claude/skills/ directories now load automatically with directory-qualified names on conflicts; the closest .claude/ directory wins for agent, workflow, and output-style resolution across nested projects. Auto mode runs a pre-launch classifier on every subagent spawn to close gaps where blocked actions could proceed without review. /doctor gets a flat tree layout with clearer icons; the workflow prompt keyword changes to purple shimmer and triggers only on explicit phrases like "run a workflow" (not any mention of the word). Remote Control connection failures now show a persistent red "/rc failed" indicator. Fixes credential-refresh races, Chrome OAuth silent failures, claude agents 401 errors on custom API gateways, ctrl+b backgrounding restarting subagents from scratch, /bg sessions stuck at "Working," and vim undo stepping.

Anthropic opens Seoul office with broad Korean AI partnerships (June 17)

Anthropic establishes its Seoul office and signs an MoU with Korea's Ministry of Science and ICT covering AI safety and cybersecurity collaboration. Enterprise deployments announced simultaneously: NAVER rolls out Claude Code to its entire engineering organization (thousands of engineers); Nexon uses Claude Code for live-service game development; LG CNS deploys Claude to thousands of employees for software development; Hanwha Solutions integrates Claude via AWS Bedrock with data-residency compliance; Samsung SDS deploys Claude across Samsung Electronics for knowledge work and agentic workflows; Channel Corp integrates Claude into Channel Talk (serving 230,000+ companies). Academic partnership with the NAIRL consortium (KAIST, Korea University, Yonsei University, POSTECH) provides 60 researchers with Claude access. Claude for Startups program launches in Korea.

Amodei at G7 (June 17): US-led AI coalition, structured frontier access, chip controls

Dario Amodei joined Google DeepMind CEO Demis Hassabis and other tech CEOs at the G7 in Évian-les-Bains for a working session with heads of state including President Trump. Amodei called for a US-led coalition to set global AI standards, proposed structured access to frontier models and chip trade that explicitly excludes China, and warned governments to "resist the temptation to splinter" on AI governance — a reference to the ongoing Fable 5/Mythos 5 export-control suspension that affects Anthropic's own newest models.

🔌 Claude Code Plugins

aws-agents-for-devsecops — new addition (June 18)

Adds AWS DevOps Agent and AWS Security Agent to Claude Code from the aws/agent-toolkit-for-aws repo: query logs, metrics, and traces across Grafana, Datadog, Splunk, New Relic, and CloudWatch for incident investigation; run code-quality reviews and release-readiness tests; scan code for vulnerabilities, run penetration tests, and apply auto-generated fixes — all via conversational prompts.

logrocket — new addition (June 18)

Connects Claude Code to LogRocket's session replay, performance metrics, and issue data via LogRocket's official remote MCP connector; surfaces user behavior, performance anomalies, and open issues through natural-language queries.

monday-crm — new addition (June 18)

Enables operating the monday CRM platform in natural language: build pipelines, generate ranked deal briefings, spin up forecast dashboards, audit board health, clean up bulk data, and convert meeting notes into deal updates — all writing back to monday as real, persisted updates.

codemod — dynamic-workflow orchestration + security hardening (June 18)

Replaces prose-based orchestration in the four main modernization commands with deterministic Workflow scripts (extract-rules.js, harden-scan.js, portfolio-assess.js, uplift-deltas.js); adds layered security hardening: untrusted code is treated as data only, second-order injection fields are marker-fenced, paths are validated against strict patterns, and a scoped scaffolder agent with minimal tool access handles all writes. Adds new /modernize-uplift command for same-stack version bumps (e.g., .NET 4.8 → .NET 8) with a leaf-first build graph and minimal-diff discipline. Older builds fall back to direct prose automatically when the Workflow tool is unavailable.

Plugin ranking cache unavailable this cycle — tracking limited to new additions and verifiable feature changes. Automated subtree pointer bumps visible in main (stripe, posthog, semgrep, cloudflare, sentry, resend, sanity, zilliz, slack, hyperframes, confidence, etc.) — no verifiable feature change in those commits.

🛠️ Skills

claude-code-system-prompts — v2.1.181 and v2.1.178 deltas

Logs two token-count swings in the window: v2.1.178 = −20,964 tokens (second-largest single-version reduction on record, tracking a large net content removal) and v2.1.181 = −3,839 tokens — both pointing to continued net trimming of Claude Code's aggregate system-prompt payload across these two releases.

Unchanged in window: anthropics/knowledge-work-plugins, glebis/claude-skills, anthropics/skills, ComposioHQ/awesome-claude-skills

🤖 Agents & Subagents

claude-agent-sdk-typescript v0.3.181 (June 17)

Adds errorCode, canUserPurchaseCredits, and hasChargeableSavedPaymentMethod to SDKRateLimitInfo — enabling apps to detect credits-required rate limits and gate purchase flows programmatically; adds tool_use_meta.icon_url to assistant messages (populated from MCP server directory metadata); fixes Remote Control sessions silently dropping file_attachments from inbound user messages.

claude-agent-sdk-typescript v0.3.179 (June 16)

Adds optional tool_use_meta sidecar to assistant messages with display-friendly tool-call names; fixes -p mode exiting before a completed background agent's notification was delivered; fixes remote stream-json sessions appearing busy during background workflows — turn results now emit at turn boundaries.

claude-agent-sdk-python v0.2.104 (June 17)

Bundles Claude CLI v2.1.181; all 30+ fixes in that release — including the network-drive write fix, prompt-cache fix on custom ANTHROPIC_BASE_URL and Foundry, macOS TUI freeze fix, and the 5-level nested-chain depth cap — land automatically on upgrade with no code changes required.

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

🔗 MCPs & Integrations

MCP Python SDK v2.0.0a2 (June 16)

Pre-release implementing full 2026-07-28 protocol types alongside per-version type packages; restructures ClientSession to operate on the JSONRPCDispatcher receive path for improved concurrency; server-initiated requests now run concurrently instead of inline in the receive loop, removing a latency bottleneck in high-throughput MCP servers. Pre-release targets a public beta on 2026-06-30 and stable v2 on 2026-07-27.

Unchanged in window: modelcontextprotocol/servers (latest release 2026.1.26, January), github/github-mcp-server (latest release v1.3.0, June 11), modelcontextprotocol/typescript-sdk (latest 2.0.0-alpha.2, April), awslabs/mcp, portainer/portainer-mcp

💡 Community — Workflows & Ideas

No relevant community posts with verified publication dates in the window.

📰 Quick Mentions

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

• Fast Company — "As G7 wraps, OpenAI and Anthropic meet with world leaders to discuss the future of AI" — June 17 wrap-up with additional context on the Évian AI governance session
• Fortune — "AI chiefs at G7 summit, token prices vs AI use" — June 18 roundup covering the G7 AI session outcomes and broader implications
• Releasebot — Anthropic Claude Code June 2026 — automated feed tracking all Claude Code releases for the month