Claude Report — 2026-05-29

🚀 Anthropic Official

Anthropic raises $65B Series H at $965B valuation (May 28) — Led by Altimeter, Dragoneer, Greenoaks, and Sequoia, the round brings Anthropic to a $965B post-money valuation with annual run-rate revenue above $47B; compute partnerships were struck with Amazon (up to 5 GW), Google/Broadcom (5 GW of TPU capacity), and SpaceX GPU access via Colossus.

Introducing Claude Opus 4.8 (May 28) — Opus 4.8 upgrades Opus 4.7 at the same price ($5/$25 per million tokens) with a new fast mode at 2.5x speed for one-third the cost, dynamic workflows that orchestrate up to 1,000 parallel subagents in Claude Code, and a Messages API change that now accepts system-role entries mid-conversation for dynamic re-steering without cache invalidation.

Anthropic opens Milan office, sixth in Europe (May 27) — The office is led by Thomas Remy as Head of Southern Europe; Italian enterprise deployments already include JAKALA (3,000+ seats), Satispay (18-month roadmap compressed to 7 months), and Bending Spoons (majority of code changes now involve Claude).

KiYoung Choi named Representative Director of Anthropic Korea (May 26) — Choi brings 30 years of technology leadership (Snowflake, Google Cloud, Adobe, Autodesk, Microsoft) ahead of a Seoul office opening; Korean Claude usage exceeds population-based expectations by more than 3.5x.

Coding agents in the social sciences (May 27) — A survey of 1,260 quantitative social scientists found only 20% regularly use coding agents despite 81% experimenting with AI chatbots; Claude Code dominates at 86% among regular users, with adoption twice as high among researchers with typically male names and 40% higher at top-tier universities.

🔌 Claude Code Plugins

Plugin tracking skipped this execution (top-10 install cache not loaded).

🛠️ Skills

anthropics/claude-plugins-official (~28.5K stars) — Most active skills repo in the window with 18+ commits (May 26–29): added vibe-prospecting plugin, Convex plugin refresh (renamed, bumped to v1.0.1), fixed hook output protocol for security-guidance, fixed Graphite workflow detection, fixed Windows Unicode decode in six git-subprocess helpers, gated XSS pattern rules to JS-family files only, and restored macOS Python 3.9 support for LLM review.

anthropics/skills (~143K stars) — One commit on May 29 from rlancemartin: added Opus 4.8 migration guide and model updates to the claude-api skill (PR #1216).

mattpocock/skills (~48.5K stars) — Four commits on May 27–28: added a new /teach command skill and refined CONTEXT-FORMAT.md rules for clarity and consistency.

Unchanged in window: obra/superpowers (~177K stars), multica-ai/andrej-karpathy-skills (~161K stars)

🤖 Agents & Subagents

wshobson/agents (~36.1K stars) — 36 commits on May 26: Copilot harness adapter merged (PR #550), plugin command skills emission, cache clearing automation, ruff formatting sweep, and Copilot documentation audit.

VoltAgent/awesome-claude-code-subagents (~20.8K stars) — README updated May 27; catalog now lists 154+ specialized Claude Code subagents across 10 categories.

VoltAgent/awesome-agent-skills (~6.9K stars) — README updated May 27; catalog covers 1,000+ agent skills from Anthropic, Google Labs, Vercel, Stripe, Cloudflare, Netlify, Sentry, Expo, Hugging Face, and Figma.

Unchanged in window: hesreallyhim/a-list-of-claude-code-agents (~1.3K stars), supatest-ai/awesome-claude-code-sub-agents (~163 stars)

🔗 MCPs & Integrations

skyvern-ai/skyvern (~21.8K stars) — Most active MCP repo in the window with 35+ commits (May 27–29): new MCP page inspection tool wiring evidence parser and page boundaries, workflow run page redesigned as compact timeline, enterprise summary cache warmer, LLM vision mode experiment, and selector + ai_fallback fields added to Action Block schema.

anthropics/anthropic-sdk-go — Released v1.46.1 on May 28: fixed BetaMessage.Accumulate dropping encrypted_content from compaction_delta streaming events, mirroring the Python SDK patch; includes regression snapshot coverage.

punkpeye/awesome-mcp-servers — ~190 community PRs merged on May 27 adding new MCP servers across finance/fintech/crypto, knowledge/memory, security, databases, and developer tools categories.

microsoft/MarkItDown (~130K stars) — Version bumped to 0.1.6 on May 26 (PR #1914).

microsoft/playwright-mcp (~33.2K stars) — Dependency update on May 28: qs bumped from 6.15.1 to 6.15.2 via dependabot.

Unchanged in window: upstash/context7 (~56.4K stars), modelcontextprotocol/servers (grammar fix only, no functional changes)

💡 Community — Workflows & Ideas

Dynamic Workflows via JavaScript orchestration (May 28) — The orchestration plan for multi-agent workflows lives in JavaScript code rather than the model's context window; intermediate results are stored in script variables, supporting up to 16 concurrent agents and 1,000 total per run, with automatic session resumption. A Bun/Zig-to-Rust rewrite of 750,000 lines at 99.8% test compatibility with dual-reviewer-per-file was cited as a reference benchmark.

Mid-conversation system messages in the Messages API (May 28) — System-role entries are now accepted mid-conversation within the messages array, enabling dynamic re-steering of long-running agentic workflows without breaking prompt caching; the minimum cacheable prompt length was also reduced from 2,048 to 1,024 tokens.

Managed Agents: Self-Hosted Sandboxes (Public Beta) (May 26, announced at Code w/ Claude London) — Tool execution runs in user-controlled infrastructure (Cloudflare, Daytona, Modal, or Vercel) while agent orchestration stays on Anthropic servers, enabling air-gapped or private-perimeter deployments without sacrificing managed orchestration.

MCP Tunnels Research Preview (May 26, announced at Code w/ Claude London) — MCP tunnels allow agents to reach private network servers without exposing them to the public internet, addressing a concrete blocker for production MCP deployments in regulated enterprise environments.

Opus 4.8 effort parameter and fast mode API changes (May 28) — The effort parameter now defaults to "high" across all surfaces; fast mode is 3x cheaper than before ($10/$50 vs $30/$150 per million tokens input/output) at 2.5x speed; a new stop_details field documents refusal responses programmatically.

📰 Quick Mentions

• anthropics/claude-code — Bug fixes shipped May 27–29: stale marketplace banner, PR badge not updating after gh pr create, Agent Teams API failures for non-ASCII names, /review GraphQL deprecation, Read tool returning partial view on token overflow, and multiple Windows rendering fixes.
• anthropics/knowledge-work-plugins — Three new plugins merged: vibe-prospecting (May 28, PR #249), Airtable integration (May 28, PR #275), and Carta cap-table/CRM/investors trio (May 27, PR #271).
• anthropics/claude-cookbooks — Updated May 28 with new notebook recipes and skills documentation covering document generation (docx, Excel, PowerPoint, PDF) examples.
• jqueryscript/awesome-claude-code — README updated May 28 (commit e434cb3c) with new skills and tools.
• Anthropic status — all incidents resolved — Four incident clusters in the window: Opus 4.8 elevated errors (May 29, resolved in 15 min), billing/subscription display errors (May 28, 38 min), two Opus 4.7 error spikes (May 27), and Claude Code in Slack error rates (May 26, resolved in ~3.5 h).