Claude Report — 2026-07-10

🚀 Anthropic Official

Claude Code v2.1.206

Ships Jul 10 with a broad batch: adds directory path suggestions to /cd, makes /doctor trim redundant checked-in CLAUDE.md content, makes /commit-push-pr auto-allow pushes to the configured push remote, adds gateway /login support for Anthropic-operated public gateway endpoints, fixes expired-login error messaging, fixes --resume/--continue keyboard input, fixes MCP request_timeout_ms being ignored, fixes CLAUDE_CODE_EXTRA_BODY being ignored by background workers, fixes an OAuth MCP re-auth loop, and improves /code-review output on claude-opus-4-8.

Claude Code v2.1.203 and v2.1.204

Ship Jul 7 and Jul 8 respectively: v2.1.203 adds a login-expiration warning for background sessions and a grey pause badge for manual permission mode, fixes a false-positive macOS low-memory regression causing 15-20s startup stalls, fixes unresponsive background sessions, fixes subagents re-running from scratch, fixes stale Windows PATH inheritance, and fixes a Bash "argument list too long" error on many-worktree repos; v2.1.204 fixes hook events not streaming during SessionStart hooks in headless sessions.

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

Python ships Jul 9 as a parity bump tracking the bundled Claude CLI version. TypeScript ships Jul 10 adding command_lifecycle frames to stream-json/SDK sessions that track uuid-stamped message states (queued/started/completed/cancelled/discarded) and fixing stale duration_api_ms reporting on zero-API results.

API key expiration

Ships Jul 8: API keys and Admin API keys can now be created with an expiration (preset, custom, or Never), Anthropic emails creators ahead of expiration for keys valid 7+ days, and an expires_at field is added to Admin API key listings.

Introducing a way to reflect on how you use Claude

Publishes Jul 9: a new "Reflect" beta (Free/Pro/Max, requires memory enabled) surfaces usage patterns — topics and timing — through the 4D AI Fluency Framework (Delegation, Description, Discernment, Diligence), excludes incognito chats and sensitive health topics, and supports quiet hours and usage reminders.

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

Announces Jul 9 that the former Federal Reserve Chair and Nobel laureate joins the LTBT, the independent governance body overseeing Anthropic's responsible-AI commitments as a Public Benefit Corporation.

Inviting hard questions

Publishes Jul 9: Anthropic solicits the public's hardest questions about AI and commits to tracking its responses publicly, drawing on surveys of 52,000 Americans and 81,000 Claude users across 159 countries via the Anthropic Institute and the Long-Term Benefit Trust.

UST is bringing Claude to physical AI

Case study published Jul 9: UST is training 20,000 employees on Claude and embedding it in semiconductor, automotive, manufacturing, healthcare-claims, telecom, and banking workflows, with human-approval controls for regulated use cases.

🔌 Claude Code Plugins

Plugin install-rank cache is unavailable this cycle — dedicated top-10 tracking is skipped per this run's instructions. No verified plugin-specific updates to report; see Quick Mentions for any incidental plugin-adjacent finds.

🛠️ Skills

Piebald-AI/claude-code-system-prompts

Logs the v2.1.206 system-prompt diff (Jul 9, +10,807 tokens), continuing its per-release token-diff tracking.

agentskillexchange/skills

Merges PR #14 (pre-landing self-review skill) and PR #15 (autonomous-session-workflow and progressive-disclosure skills) on Jul 9, then runs two more registry sync cycles on Jul 10 pulling newly published skills from the exchange.

JSONbored/awesome-claude

Adds five new content entries (Prompt flow, Unsloth, LocalAI, Mem0, Crawl4AI) and ships web features — a freshness-distribution panel, a prerequisite-readiness checklist, and compare-tray analytics normalization — across roughly 20 commits on Jul 10.

kpab/claude-fable-5-skills

Merges PR #3 on Jul 10, bumping to v1.1.1 and adding an "API shape" section to its effort-calibrator skill.

Unchanged in window: alirezarezvani/claude-skills (PRs #906/#907 still open/draft), glebis/claude-skills, anthropics/skills, ComposioHQ/awesome-claude-skills.

🤖 Agents & Subagents

anthropics/claude-agent-sdk-typescript

Ships v0.3.206 (Jul 10), the same release covered under Anthropic Official above, adding command-lifecycle tracking frames to SDK sessions.

Unchanged in window: nicobailon/pi-subagents (still v0.34.0, Jul 7), 0xmmo/crew (latest commits dated Jul 4, no activity this window).

🔗 MCPs & Integrations

ChromeDevTools/chrome-devtools-mcp

Merges roughly 14 commits across Jul 9-10 — heap-snapshot filtering/aggregates, a context/page-management refactor moving the DevTools universe to McpPage, a telemetry fix, and exact bundled-package-version reporting during release — and becomes GitHub's top-trending MCP repo this week (+1,521 stars, now 46.6k total).

modelcontextprotocol/typescript-sdk

Ships a full 2.0.0-beta.3 release wave across eight packages (server, node, hono, fastify, express, core, codemod, client) on Jul 9, covering auth/OAuth discovery fixes, content-type validation, and cross-bundle error handling.

iiTzSenn/Sens

Ships three more releases in one day — v0.6.0, v0.7.0, and v0.8.0 (all Jul 9) — accelerating its cadence further.

shlokkhemani/rabbithole

Finishes its six-phase build arc on Jul 9: ships a canvas-first v0.4 redesign, a single Pages deploy separating site from app, portable .rabbithole files with evals and real authoring, a home-screen design pass, and browser-based PDF/URL/arXiv ingestion via a fetch-proxy Worker.

DeusData/codebase-memory-mcp

Merges PRs #987, #989, and #990 plus a CI regex fix on Jul 9, shifting from feature work into CI/tooling hardening.

Unchanged in window: awslabs/mcp (one minor Jul 9 path-validation patch, #3807, no packaged release), modelcontextprotocol/python-sdk (still v2.0.0b1, Jun 30).

💡 Community — Workflows & Ideas

addyosmani/agent-skills

"Production-grade engineering skills for AI coding agents" collection trends Jul 10 with 76,411 stars, gaining roughly 2,554 in a single day.

pilotfish

Ships v1.1.2 "hard leaf agents" (Jul 10, 304 stars), expanding its Claude-plans/cheap-workers delegation pattern to six role-based agents — scout, explore, mech-executor, executor, verifier, security-executor — each with fresh-context verification.

guizang-material-illustration

Updates Jul 8 (509 stars): a Claude Code skill for generating annotated illustration graphics, chart beautification, and reference images.

claude-codex-battery

Ships v1.3.1 (Jul 9, 72 stars): a macOS menu-bar widget showing Claude Code and Codex usage limits as battery icons, pulling data via Anthropic's OAuth API and local Codex session logs.

Danus

Updates Jul 10 (50 stars): a Python tool orchestrating mathematical-reasoning agents around a "fact-graph" memory structure for Claude-driven math problem solving.

📰 Quick Mentions

• ditto — v0.1.2 (Jul 8) mines Claude Code logs into local agent profiles; roadmap teases a one-command installer next.
• cc-vision-hook — v0.2.0 (Jul 9), a hook toolkit giving vision-blind models text descriptions of pasted or tool-produced images.
• claude-paper-tools — new Jul 10 (24 stars), Claude Code skills for academic paper analysis including GRADE evidence-quality appraisal.
• wechat-styler — updated Jul 8 (19 stars), a Claude Code skill converting Markdown into WeChat-styled HTML.
• search-mesh — new Jul 10 (3 stars), a Rust SIMD-accelerated codebase-intelligence MCP server for AI agents.
• dali-mcp — new around Jul 9 (5 stars), an MCP server that scores a prompt before it spends the credit.
• grokbuild-proxy — new Jul 10 (20 stars), a self-hosted Go proxy offering Grok-Build-style compatibility alongside Claude Code tooling.