CLAUDE·news
PTEN
№ 0602026-06-11

Claude Report — 2026-06-11

🚀 Anthropic Official

Claude Code v2.1.172 — Nested Sub-Agents (5 Levels Deep) + Fixes

June 10. Sub-agents can now spawn their own sub-agents, capped at 5 levels deep — reversing a two-year hard constraint and enabling context-delegation hierarchies where noisy work is offloaded to throwable child windows. Also: Bedrock reads AWS region from ~/.aws config when AWS_REGION isn't set; /plugin gains a search bar for marketplace browsing; model attribute added to the claude_code.lines_of_code.count OTEL metric; fixes for 1M-context sessions stuck without credits, repeating image-processing error, agents view stuck on "Working" for 30+ seconds, WebFetch wildcard subdomain rule matching, and team memory recall in remote sessions; workflow validation no longer rejects scripts that mention Date.now() or Math.random().

Claude Code v2.1.169 — Safe Mode, /cd, disableBundledSkills

June 8. Adds --safe-mode flag and CLAUDE_CODE_SAFE_MODE env var to disable all customizations for troubleshooting sessions. Adds /cd command to move a session to a new working directory without invalidating the prompt cache. Adds disableBundledSkills setting and CLAUDE_CODE_DISABLE_BUNDLED_SKILLS env var to suppress built-in skills. Also fixes Up/Down arrows skipping wrapped rows in long inputs, enterprise managed MCP policies not enforcing on reconnect or IDE-typed configs, a ~30–50ms UI stall on macOS at turn start, and claude -p slowness on Windows during slash-command scan; improves Remote Control, Git Credential Manager, and background agent reliability; reduces idle CPU usage.

Claude Code v2.1.173 — Fable 5 Suffix Fix

June 11. Strips the [1m] suffix that appeared appended to some Fable 5 model-name strings so they normalize correctly; removes a spurious "sandbox dependencies missing" warning on Windows at startup.

Claude Managed Agents — Scheduled Deployments + Env-Var Credential Vaults (Public Beta)

June 9. Sessions can now run on a cron schedule: each fire starts a fresh session, completes the task, then exits; customers can pause, resume, archive, or trigger ad-hoc runs from the Console or API. Vault now supports environment variable credentials — the real secret is injected at the network boundary only on requests to allowlisted domains; the agent sandbox never sees it, blocking prompt-injection–based secret exfiltration. session.thread_* webhook events also gain a new session_thread_id field identifying the multi-agent thread that triggered the event.

ClaudeForFoundationModels — Swift Package for Apple Foundation Models (WWDC 2026)

June 8. New beta Swift package conforms Claude to Apple's LanguageModel protocol so developers drive it with the same LanguageModelSession API used for the on-device model: respond(to:), streaming, guided generation (@Generable), client-side tools, and server-side tools (web search, web fetch, code execution) all work identically. Requests go app→Claude API directly; Apple is not in the request path. Supports .opus4_8, .sonnet4_6, .fable5, and other ClaudeModel constants; fixedEffort: controls reasoning depth per model. Targets iOS 27, macOS 27, visionOS 27, watchOS 27 betas; marked beta with potential API changes before general availability.

Connector Observability Dashboard + In-App Directory Submission (Public Beta)

June 8. Connector owners on Team or Enterprise now have a dashboard tracking active users, total tool calls, directory ranking over time, a composite health score, per-product error rates, and latency. MCP servers can now be submitted directly to the Anthropic Connectors Directory from within Claude (Organization Settings → Directory), replacing the external submission flow; requires Admin/Owner access or a custom role with the Directory management permission.

API: New reasoning_extraction Refusal Category on Claude Fable 5

June 9. stop_details.category on Fable 5 now returns "reasoning_extraction" when a request is blocked for attempting to reverse-engineer or duplicate model outputs under Anthropic's ToS. No beta header required; existing "cyber" and "bio" categories are unchanged.

🔌 Claude Code Plugins

Plugin cache unavailable — top-10 tracking skipped this run.

🛠️ Skills

None of the top 5 had verifiable updates in the window.

Unchanged in window: anthropics/knowledge-work-plugins (Jun 11 commit activity detected, details unavailable), glebis/claude-skills, anthropics/skills, Piebald-AI/claude-code-system-prompts, ComposioHQ/awesome-claude-skills

🤖 Agents & Subagents

claude-agent-sdk-typescript v0.3.172 — skipMcpDiscovery per Plugin

June 10. The plugins option now accepts skipMcpDiscovery: true per entry, letting plugin hosts that manage a plugin's MCP connections load skills/hooks from the plugin path without the engine re-reading its .mcp.json. Also fixes slash-followed-by-whitespace input (e.g., / add tests) being silently dropped instead of treated as a plain prompt.

claude-agent-sdk-typescript v0.3.169 — Experimental Session Usage API + Browser SSE

June 8. Adds an experimental usage_EXPERIMENTAL_MAY_CHANGE_DO_NOT_RELY_ON_THIS_API_YET() method on Query returning structured session cost, plan rate-limit, and local usage-behaviors data — useful for cost dashboards before the API stabilizes. Also adds an sse option (SSEOptions) to BrowserQueryOptions as an alternative to WebSocket for browser SDK consumers.

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

🔗 MCPs & Integrations

github/github-mcp-server v1.2.0

June 8. Aligns with the 2026-01-26 stable MCP specification. Adds create_project and create_iteration_field to the projects write API; restores thread ID in pull request review comment responses; adds include_patch parameter for commit retrieval; adds confidence parameter for issue mutations; fixes project item ID usability and assignees-array clearing; adds installation guides for Zed and OpenCode. 12 contributors (7 making their first contribution).

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

💡 Community — Workflows & Ideas

claudefa.st: "Claude Code Nested Subagents: 5 Levels Deep"

Published Jun 10–11. Practical breakdown of the new v2.1.172 nesting capability: covers the context-management motivation (offload noisy work to throwable child windows to keep parent context lean), the depth ceiling, token-cost math for multi-level chains, and the key heuristic for when to delegate versus stay flat.

YouTube: "Claude Code v2.1.172 — Agents that spawn agents"

Published Jun 10. Video walkthrough demonstrating a 3-level nested agent chain on a large codebase scan, showing how the top-level agent's context stays clean while child agents handle log greps and web searches.

📰 Quick Mentions

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

← back to archiveend of edition № 060