CLAUDE·news
PTEN
№ 0532026-06-04

Claude Report — 2026-06-04

🚀 Anthropic Official

Claude Code v2.1.162 — waitingFor, Native Grep/Glob, Devin Desktop Rename

Released June 3, 2026. Adds waitingFor to claude agents --json (shows what a session is blocked on, e.g. a permission prompt); explicitly naming Grep/Glob in --tools now provisions the fast embedded search tools on native builds; /effort confirms when a level persists as the default; autocomplete fills slash commands into the prompt instead of running immediately; Remote Control switches from a startup message to a persistent footer pill; Windsurf is renamed Devin Desktop in /ide, /terminal-setup, and /scroll-speed. Fixes silent startup hang on read-only config dirs, WebFetch permission rules for preapproved domains, Windows backslash path matching, dropped Esc interrupts in SDK sessions, emoji-near-truncation 400 errors, and MCP sub-1 s timeout flooring.

Claude Code v2.1.161 — OTEL Labels, Agent Progress Counts, MCP Collapse

Released June 2, 2026. OTEL_RESOURCE_ATTRIBUTES values now appear as labels on metric datapoints for custom dimensions; claude agents rows show done/total counts for fanned-out work; /mcp collapses unused claude.ai connectors behind "Show unused connectors"; failed Bash commands no longer cancel other parallel tool calls; Linux fullscreen clipboard uses wl-copy/xclip/xsel. Fixes: managed-settings policies blocking third-party sessions, background subagent stdout corruption, /mcp printing secrets/credentials to terminal, and LSP workspaceSymbol returning empty results.

API (Jun 2): Advisor max_tokens Cap + Zero-Output Refusals Unbilled

The advisor tool now accepts a max_tokens parameter to cap per-call output length, reducing latency and token cost for workloads that don't need full-length advisor responses. Separately, requests returning stop_reason: "refusal" with zero generated output are no longer billed.

What We Learned Mapping a Year's Worth of AI-Enabled Cyber Threats

Published June 3, 2026. Analysis of 832 banned accounts engaged in malicious AI-assisted attacks (Mar 2025–Mar 2026): medium-risk-or-higher actors rose from 33% to 56% (1.7×); 67.3% used AI for malware writing vs. 6.5% for lateral movement; a state-sponsored Nov 2025 campaign scored 100/100 risk across 30 MITRE techniques in 13 tactics. Anthropic releases an interactive attack-visualization tool and flags ATT&CK framework gaps for agentic orchestration behaviors.

Lessons from Building Claude Code: How We Use Skills

Published June 3, 2026 by Thariq Shihipar (MTS, Anthropic). Documents 9 internal skill categories (library/API reference, product verification, data fetching, business automation, code scaffolding, code quality, CI/CD, runbooks, infra ops) plus design patterns: SKILL.md progressive disclosure, gotchas sections for common failure points, model-focused trigger descriptions as discovery keywords, and append-only logs for cross-session continuity.

Running an AI-Native Engineering Org

Published June 3, 2026 by Fiona Fung (Director of Engineering, Claude Code and Cowork). Planning moves to just-in-time prototyping; human code review concentrates on security, legal, and product judgment while Claude handles style, linting, and bug detection; verification becomes the new development bottleneck; hiring prioritizes creative builders with product sense over raw coding throughput.

🔌 Claude Code Plugins

Plugin cache unavailable — top-10 tracking skipped this run. No new plugin updates confirmed in the June 3–4 window beyond yesterday's coverage.

🛠️ Skills

Piebald-AI/claude-code-system-prompts — Synced to v2.1.162

Synced June 3, 2026. Adds +9,871 tokens covering new slash-command instructions for React design systems, broader AI coding tool configuration discovery (additional rule directories scanned), and expanded LSP and notebook editing tool descriptions.

Unchanged in window: anthropics/knowledge-work-plugins, VILA-Lab/Dive-into-Claude-Code, ComposioHQ/awesome-claude-skills, anthropics/skills

🤖 Agents & Subagents

claude-agent-sdk-typescript v0.3.162 — Refusal stop_reason + Fast Search Default

Released June 3, 2026. Refusal error messages now carry stop_reason: "refusal" and stop_details on the assistant message and in session transcripts, enabling programmatic detection without text-matching. Sessions on native builds default to fast embedded find/grep instead of always registering dedicated Grep/Glob tools; opt back in by naming them in tools or allowedTools.

claude-agent-sdk-typescript v0.3.161 — Idempotent init, Live Agent Switching

Released June 2, 2026. initialize is now idempotent — a second call returns the same success payload instead of Already initialized. ControlResponse gains an optional pending_permission_requests field; applyFlagSettings live-applies agent changes so switching the active agent (or passing null to reset) takes effect on the next turn without session restart.

claude-agent-sdk-python v0.2.89 — CLI 2.1.162 Bundle

Released June 3, 2026. Bundles Claude CLI v2.1.162; no SDK-level API changes in this release.

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

🔗 MCPs & Integrations

No new MCP server releases confirmed in window.

Unchanged in window: modelcontextprotocol/servers, github-mcp-server, awslabs/mcp, modelcontextprotocol/typescript-sdk, portainer/portainer-mcp

💡 Community — Workflows & Ideas

Best Practices for Getting Started with Claude Cowork

Published June 3, 2026. Team adoption playbook: start with async document reviews before live collaboration, maintain a shared project CLAUDE.md as the team's single source of truth, and run /team-onboarding to package your setup for new members — mirrors the individual-developer pattern of layering skills on top of Claude Code defaults.

How Anthropic Enables Self-Service Data Analytics with Claude

Published June 3, 2026. Describes Anthropic's internal analytics runbook skill: pre-loads the warehouse schema and common query patterns into context so non-technical users get ad-hoc answers without a data engineer — a concrete production example of the "data fetching and analysis" skill category documented in the companion engineering post.

📰 Quick Mentions

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

← back to archiveend of edition № 053