Anthropic News Flash | Claude Code Plugin Enhancements and Auto Mode Expansion

On the morning of June 1, 2026, the Clauder Navi editorial team brings you the latest official Anthropic news in a broadcast style. The end of May was a quiet period with no new articles from the News or Engineering Blog, but Claude Code received two consecutive updates, and an official annual survey report was also published.

Today we cover four stories: [1] Plugin enhancements in Claude Code v2.1.157, [2] Auto mode expansion in v2.1.158, [3] the release of the first annual survey report, and [4] the Agent SDK billing separation arriving in two weeks.

Article Summary by AI Chatpowered by Claude
結論powered by Claude

The end of May 2026 at Anthropic saw zero new articles from the News or Engineering Blog, yet it was a fruitful period as Claude Code received two consecutive updates. The standout feature of v2.1.157 is support for automatic loading of plugins placed in the .claude/skills directory, which significantly reduces the effort required to distribute custom in-house skills.

The following v2.1.158 brought Auto mode support for Opus 4.7 / 4.8 across three platforms — Bedrock, Vertex AI, and Foundry — opening up long-running automated task execution to teams deploying Claude Code via cloud infrastructure. The "2026 Agentic Coding Trends Report" published around the same time quantified what the report calls the "delegation gap" as an annual survey finding: developers involve AI in roughly 60% of their work, yet only about 20% of tasks feel safe to delegate fully.

Not to be overlooked is the Agent SDK billing separation taking effect on June 15. With fewer than two weeks remaining, developers who have embedded the Agent SDK or claude -p in their scripts need to review their usage in advance. Guidance on how to handle this is covered in a previously published dedicated guide.

目次 (6)

[1] Claude Code v2.1.157 Adds Automatic Skills Loading and Plugin Enhancements

Claude Code v2.1.157 was released at 20:20 JST on May 29, 2026. The biggest change is automatic loading of skills placed in the .claude/skills directory. Skill definitions previously distributed as plugins can now be loaded simply by placing them under .claude/skills, without needing to register them in a marketplace. A new claude plugin init <name> command was also added to scaffold new plugins. Additionally, the claude agents command now reflects the agent field in the configuration file (overridable with --agent <name>), and EnterWorktree allows switching to a different managed worktree during a session. Bug fixes span more than 37 items covering MCP, background sessions, terminal UI, and IDE integrations (VS Code / Cursor / Windsurf).

Impact for engineer readers: ★★★. With automatic .claude/skills loading, distributing standardized in-house skills no longer requires going through a marketplace — simply placing them in the .claude/ directory makes them available to the entire team. Numerous IDE integration fixes make an immediate update recommended for users working with VS Code / Cursor / Windsurf. Source: claude-code v2.1.157 release notes.

[2] Claude Code v2.1.158 Brings Auto Mode Support for Bedrock / Vertex / Foundry

Claude Code v2.1.158 was released at 02:42 JST on May 30, 2026. The single change in this release is that Auto mode is now available with Opus 4.7 and Opus 4.8 on Bedrock, Vertex AI, and Foundry. Activation requires setting the environment variable CLAUDE_CODE_ENABLE_AUTO_MODE=1 as an opt-in. Auto mode is an execution mode that lets the model proceed through long-running tasks on its own judgment; support has now expanded from a limited set of paths to three cloud integrations.

Impact for engineer readers: ★★. For teams deploying Claude Code via AWS Bedrock or Google Vertex AI, being able to try Auto mode for long-running tasks while retaining their existing cloud billing and authentication infrastructure is a practically valuable change. The ease of switching via a single environment variable also lowers the barrier from testing to production deployment. Source: claude-code v2.1.158 release notes.

Anthropic has released an annual survey report for developers titled "2026 Agentic Coding Trends Report." The central argument is the "delegation gap": despite developers involving AI in roughly 60% of their work, the share of tasks they feel comfortable fully delegating remains at just 0–20%. The report categorizes eight trends under three pillars — "Foundation," "Capabilities," and "Impact" — and includes case studies from Rakuten, CRED, TELUS, and Zapier. It is structured around three core themes: multi-agent coordination, human-AI collaboration patterns, and expansion to non-technical teams.

Impact for engineer readers: ★★. The concept of the "delegation gap" speaks directly to a common challenge in evaluating the ROI of AI tool adoption. For Japanese organizations, the inclusion of a Rakuten case study makes this a useful reference for engineering managers and technical leads building their adoption roadmaps. Source: 2026 Agentic Coding Trends Report.

[4] Agent SDK Billing Separation Set for June 15 — Less Than Two Weeks Away

The Agent SDK billing separation announced in official documentation on May 13, 2026, is now fewer than two weeks away from its June 15 effective date. After it takes effect, usage via the Agent SDK, claude -p, and third-party agent tools will be separated from the conversational quota of subscriptions and managed under dedicated monthly credit allocations based on each plan — Pro ($20/month) / Max 5x ($100/month) / Max 20x ($200/month). The monthly subscription fee does not directly translate into the credit amount; each plan has a separate credit pool. Unused credits expire at the end of the month and do not roll over. Meanwhile, interactive Claude Code (terminal / IDE), claude.ai web / desktop / mobile, and Cowork will continue to operate under the existing subscription quota. Details on what happens when the credit limit is exceeded are expected to be communicated by email around June 8.

Impact for engineer readers: ★★★. If you have embedded claude -p in production scripts or continuous processing workflows, you need to cross-reference past usage against your credit amount before June 15. A seven-step guide covering path determination, cost estimation, and internal escalation criteria is available at Claude Agent SDK Billing Separation on 6/15 | Impact on Pro/Max and How to Respond. Sources: Agent SDK overview, Using the Claude Agent SDK with your Claude plan.

What to Watch Next

The end of May was a quiet period with no new News or Engineering Blog articles, but on the tooling side, back-to-back Claude Code updates and the publication of an official annual survey marked steady progress on practical foundations. The biggest milestone in June is undoubtedly the Agent SDK billing separation on June 15. Details on handling cases where the credit limit is exceeded are expected around June 8, and teams with automated workflows integrated into their operations would be wise to wait for that guidance before making final adjustments. It will also be worth watching how widely the plugin auto-loading introduced in v2.1.157 takes hold as a standard in team development. The Clauder Navi editorial team will continue to bring you the latest from Anthropic's official channels.

Sources

参考になったら ♡
Clauder Navi 編集部
@clauder_navi

Anthropic の Claude / Claude Code を中心に、日本のエンジニア向けに最新動向と実務 を毎日発信。 運営方針 は メディアについて をご覧ください。