Claude Report — 2026-05-30

🚀 Anthropic Official

• Claude Code v2.1.158 — Enables Auto mode on Amazon Bedrock, Vertex AI, and Microsoft Foundry for Opus 4.7 and Opus 4.8 via CLAUDE_CODE_ENABLE_AUTO_MODE=1, extending the complexity-based routing mechanism beyond the first-party API to all three major cloud-hosted deployment surfaces.

• Claude Code v2.1.157 — Introduces auto-loading of plugins placed in .claude/skills directories without marketplace registration, adds claude plugin init <name> to scaffold new plugins in place, improves /plugin argument autocomplete, and adds mid-session worktree switching via EnterWorktree.

• Claude Code v2.1.156 — Fixes a bug where Opus 4.8 thinking blocks were being modified during processing, causing API errors in high-effort inference sessions.

• Claude Managed Agents on AWS (GA) — Releases webhooks for session/vault lifecycle events, multiagent orchestration, and self-hosted sandboxes to general availability on Claude Platform on AWS, introducing IAM authentication and the AnthropicSelfHostedEnvironmentAccess managed policy for private-network agent deployments.

• Claude API — Opus 4.8 launch features — Adds mid-conversation system messages (role: "system" at non-first positions) that preserve prompt cache hits when instructions change mid-session, and returns stop_details.refusal_category so applications can route policy, safety, and capability refusals to separate fallback paths.

• anthropic-sdk-python v0.105.0 — Adds claude-opus-4-8 model support, mid-conversation system blocks, usage.output_tokens_details, custom file size caps, and renames managed-agents private-sandbox-worker to self-hosted-sandbox-worker in examples.

• anthropic-sdk-python v0.105.1 — Switches PyPI releases to Trusted Publishing, improving supply-chain security for the CI/publishing pipeline.

• anthropic-sdk-typescript sdk v0.100.0 — Adds claude-opus-4-8, mid-conversation system blocks, and usage.output_tokens_details to the TypeScript SDK, matching the Python SDK's Opus 4.8 parity.

• anthropic-sdk-typescript sdk v0.100.1 — Fixes streaming to carry encrypted_content on beta compaction blocks, and updates lockfiles for standardwebhooks dependencies.

• anthropic-sdk-typescript sdk v0.99.0 — Adds custom file size cap support and fixes streaming to carry stop_details through message_delta accumulation.

• anthropic-sdk-go v1.46.0 — Adds claude-opus-4-8, mid-conversation system blocks, usage.output_tokens_details, custom file size caps, and renames the private-sandbox-worker example to self-hosted-sandbox-worker.

• claude-agent-sdk-typescript v0.3.158 — Reaches parity with Claude Code v2.1.158, enabling Auto mode on Bedrock, Vertex, and Foundry for Opus 4.7 and 4.8 in the TypeScript Agent SDK.

• claude-agent-sdk-typescript v0.3.157 — Reaches parity with Claude Code v2.1.157, bringing plugin auto-loading, scaffolding, and worktree/session fixes to the TypeScript Agent SDK.

• claude-agent-sdk-typescript v0.3.156 — Reaches parity with Claude Code v2.1.156, applying the Opus 4.8 thinking block fix to the TypeScript Agent SDK.

• claude-agent-sdk-typescript v0.3.154 — Fixes stdio MCP servers being incorrectly restarted on every reconcile pass due to config-equality false positives.

• claude-agent-sdk-typescript v0.3.152 — Adds reloadSkills: true return support in SessionStart hooks, sessionTitle via hookSpecificOutput, and a new MessageDisplay hook event for transforming or suppressing assistant message text during display.

🔌 Claude Code Plugins

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

🛠️ Skills

• knowledge-work-plugins — Four new marketplace plugins — Merges desktop-commander (MCP server for terminal commands, file operations, and multi-format document handling), qodo-skills (pre-commit code review catching issues before they land), servicenow-sdk (create and deploy ServiceNow apps using the Fluent SDK), and twilio-developer-kit (procedural knowledge for SMS, Voice, WhatsApp, Verify, and 30+ Twilio products).

Unchanged in window: mattpocock/skills, multica-ai/andrej-karpathy-skills, anthropics/claude-plugins-official, anthropics/skills

🤖 Agents & Subagents

• wshobson/agents — Native plugin-install for Codex, Cursor, and Gemini — Adds native plugin install support for Codex, Cursor, and Gemini environments alongside Claude Code, and introduces a CLAUDE.md to AGENTS.md symlink to unify agent configuration across harnesses.

• obra/superpowers — Mandatory contributor self-identification — Requires all PR and issue authors to disclose their model, harness, harness version, and installed plugins, and mandates all PRs target the dev branch rather than main.

Unchanged in window: VoltAgent/awesome-claude-code-subagents, VoltAgent/awesome-agent-skills, hesreallyhim/a-list-of-claude-code-agents, supatest-ai/awesome-claude-code-sub-agents

🔗 MCPs & Integrations

• modelcontextprotocol/servers — Filesystem server dollar-sign fix — Corrects literal dollar-sign handling in the filesystem MCP server to prevent unintended variable substitution during path and string operations.

• skyvern-ai/skyvern — May 30 burst (9 commits) — Advances the runtime verification evidence tracking arc (PRs 10-13), adds sidebar flyout hover reveal, fixes browser session UI refresh after creation, and continues the anti-bot blocker rendering work.

Unchanged in window: punkpeye/awesome-mcp-servers, microsoft/playwright-mcp, upstash/context7, microsoft/MarkItDown

💡 Community — Workflows & Ideas

• Dynamic Workflows internals (HN discussion) — Explains that Claude writes a JavaScript orchestration script at runtime executed by a separate background process, keeping the workflow plan outside the model's context window to avoid token-accumulation bottlenecks; scripts must be deterministic (no Math.random()) to support resumable execution.

• Ask HN: Dynamic Workflows vs. session handoff patterns — Community clarifies that Dynamic Workflows target background multi-agent fan-out while skills like claude-handoff-revive address single-session context preservation across interruptions, confirming the two patterns serve orthogonal failure modes.

• Show HN: Claude Code Workflow (CCW) — Spec-driven pipeline generator — Introduces an open-source tool that lets developers define an entire multi-step agentic pipeline in a single YAML file, then auto-generates all plugin manifests, slash commands, and install scripts from that single source of truth.

• claude-cookbooks — Agent SDK Kubernetes hosting follow-up — Moves AGENT_IMAGE validation to startup for graceful readiness probe failure, improves tenant token loading with explicit validation and logging, adds comments on body-size limits and race conditions, and fixes cluster reuse workflow in kind-quickstart.sh.

📰 Quick Mentions

• anthropic-sdk-python v0.105.2 — Minor follow-up granting id-token permission for PyPI Trusted Publishing in CI.
• anthropic-sdk-typescript foundry-sdk v0.2.4 — Updates lockfiles to include proper standardwebhooks dependencies.
• anthropic-sdk-typescript aws-sdk v0.3.1 — Updates lockfiles to include proper standardwebhooks dependencies.
• EveryInc/compound-engineering-plugin v3.9.3 — Fixes routing of explicit external-research requests by intent, and corrects emission of repo root path instead of basename in subshell context.
• quemsah/awesome-claude-plugins — Data refresh 28.05.2026 — Automated dataset refresh indexes 19,041 total repositories in this Claude Code plugin adoption tracker, maintained via n8n workflows.