CLAUDE·news
PTEN
№ 0902026-07-12

Claude Report — 2026-07-12

🚀 Anthropic Official

Claude Code v2.1.206

Ships Jul 10: adds directory-path suggestions to /cd, a /doctor check for trimming checked-in CLAUDE.md files, and auto-allows pushes to the configured push remote from /commit-push-pr; adds gateway /login support for Anthropic public gateway endpoints and lets background agents upgrade silently after a Claude Code update; fixes MCP servers ignoring per-server request_timeout_ms, OAuth MCP servers requiring re-authentication after token refresh, the /model picker showing prices for the wrong model, Remote Control task-status updates being lost during connection recovery, Deep Research mislabeling Fetch-phase agents as "unknown," and Bedrock repeatedly re-requesting AWS SSO credentials.

A new way to reflect on how you use Claude

Launches Jul 9 in beta for Free/Pro/Max users with memory on: a dashboard (Settings → Reflect) showing 1/3/6/12-month usage recaps by topic and task type, with quiet-hours and break-reminder controls and prompts asking what users still want to do themselves; excludes incognito chats, connected-tool source files, and health-integration conversations.

Inviting hard questions

Launches Jul 9: a public campaign asking users to submit the toughest questions about AI's societal impact, building on a 52,000-person US survey and an 81,000-user Claude poll across 159 countries; Anthropic commits to publishing responses and progress reports.

Ben Bernanke appointed to Anthropic's Long-Term Benefit Trust

Announces Jul 9: former Fed Chair and 2022 Nobel laureate Ben Bernanke joins the independent trust (alongside Neil Buddy Shah, Richard Fontaine, and Mariano-Florentino Cuéllar), lending economic-research expertise; trustees hold no equity and are compensated only for time served.

UST is bringing Claude to physical AI

Publishes Jul 9: UST embeds Claude into engineering platforms for semiconductor, automotive, and manufacturing clients via its UST-iDEC platform — Claude Code reads chip pinouts/hardware schematics and writes regression tests, cutting hardware-validation cycle times 50-70%; UST commits to training 20,000 employees on Claude.

Claude Agent SDK Python v0.2.115 and TypeScript v0.3.206

Python v0.2.115 (Jul 9) updates the bundled CLI to v2.1.206; TypeScript v0.3.206 (Jul 10) adds command_lifecycle frames to stream-json output and SDK sessions, giving consumers visibility into command start/stop events mid-stream.

🔌 Claude Code Plugins

Plugin install-rank cache is unavailable this cycle — top-10 tracking is skipped per this run's instructions. The only activity in anthropics/claude-plugins-official this window was routine automated SHA-pin dependency bumps in marketplace.json (dozens of bump(...) commits Jul 10-12), not substantive plugin news.

🛠️ Skills

JSONbored/awesome-claude

Merges two new feature PRs on Jul 12 — a theme-distribution panel for the browse view (#4881) and a safety/privacy surface panel on entry detail pages (#4877) — following a run of analytics-instrumentation PRs the prior day.

Unchanged in window: Piebald-AI/claude-code-system-prompts (latest diff already logged for v2.1.207 pre-window), agentskillexchange/skills (only automated regen/sync bot commits), kpab/claude-fable-5-skills, alirezarezvani/claude-skills, glebis/claude-skills, anthropics/skills, ComposioHQ/awesome-claude-skills.

🤖 Agents & Subagents

nicobailon/pi-subagents

Ships Jul 10: adds configurable agent-launch defaults and hides shadowed agents from list output, plus fixes for resolving the Pi CLI inside embedded Windows hosts and a misleading "now ago" activity-status string.

anthropics/claude-agent-sdk-typescript

Ships v0.3.206 (Jul 10), covered under Anthropic Official above, adding command_lifecycle stream frames for tracking command execution state.

Unchanged in window: 0xmmo/crew (latest commit Jul 4).

🔗 MCPs & Integrations

shlokkhemani/rabbithole

Pushes Phase 9 work on Jul 11 — audit-correctness and security fixes (Slice 1), single-journey and temporal test coverage (Slice 2), and hygiene/governance closure (Slice 3) — continuing directly from last window's Phase 8 close-out.

DeusData/codebase-memory-mcp

Merges a variable-length-path fix (PR #1036) and shortest-path var-length Cypher semantics on Jul 11, plus coalesce-in-WHERE support (PR #1034) and sequential JSON/branch-file restoration (PR #1030); Jul 10 added TOON compact output for search_graph (~90% fewer response tokens) and hook-based JSON-format requests for internal graph searches.

awslabs/mcp

Adds a list-invoice-summaries tool to the billing-cost-management MCP server for real invoice data with FX and tax breakdowns (Jul 9), alongside path-validation hardening for params_file reads and continued Redshift/mssql/oracle server work carried over from Jul 10.

modelcontextprotocol/typescript-sdk 2.0.0-beta.3

Ships Jul 9: adds runtime-neutral Bearer authentication for web-standard fetch hosts (Cloudflare Workers, Deno) via new verifyBearerToken/bearerAuthChallengeResponse exports, Standard Schema (Zod) support in elicit(), HTTP content-type validation with 415 responses, and cross-bundle instanceof fixes for SDK error classes.

iiTzSenn/Sens v0.8.0

Ships Jul 9: adds "dynamic command suggestions" that append context-aware next-step commands (e.g. sens who login) to every query result, operating entirely at the terminal level without touching the model or MCP protocol.

Unchanged in window: ChromeDevTools/chrome-devtools-mcp (latest v1.5.0, Jul 3), modelcontextprotocol/python-sdk (still v2.0.0b1, Jun 30).

💡 Community — Workflows & Ideas

Skillgrade v0.2.1

Ships Jul 11: increases the API request timeout to 10 seconds and adds configurable model environment variables with dynamic API model resolution, extending the YAML-based skill-evaluation framework covered in yesterday's report.

📰 Quick Mentions

← back to archiveend of edition № 090