Claude Report — 2026-10-23

🚀 Anthropic Official

Claude Code v2.1.309 (Oct 22)

• Fixes a notable batch of sandbox/security issues: sandboxed Bash on macOS/Linux was running shell startup files outside the sandbox before each command; macOS denyWrite glob patterns weren't applying through symlinks; sandboxed commands/auto-accepted edits could modify the MCP config file; a disabled MCP server could get silently reconnected by a later permission prompt.
• Fixes worktree-isolated sessions wrongly accepting locale-prefixed and arithmetic commands, Bash permission checks skipping prompts for compound directory-changing commands, and Edit(path) deny rules not blocking Bash redirects.
• Changes sed -i to run with the standard prompt instead of a simulated preview, makes claude ultrareview require workspace trust, and makes dangerous rm/rmdir under bypass-permissions deny immediately with a rewrite hint.
• Updates the bundled dataviz skill with an APCA text-contrast measure and the /claude-api skill for service-account keys and Workload Identity Federation token management.
• Fixes numerous VS Code, PowerShell, Windows, and cloud-session papercuts (folding of long slash-command output, Ctrl+J behavior, Git for Windows checks, gateway keepalive flicker, etc.).

Claude Agent SDK (TypeScript) v0.3.309 / Python v0.2.173 (Oct 22)

Parity releases tracking Claude Code v2.1.309; TypeScript additionally fixes managedSettings silently dropping certain properties and tightens temp-directory handling for the extracted CLI binary.

Donating Agent Skills to the Agentic AI Foundation (Oct 22)

Transfers the Agent Skills standard — name, logo, GitHub org, and the agentskills.io domain — to the Agentic AI Foundation (AAIF), the same Linux Foundation body that already governs MCP, co-founded by Anthropic, Block, and OpenAI; cites 70+ integrating platforms (ChatGPT, Cursor, GitHub Copilot, VS Code), ~1M skills published, and ~280M installs.

LLMs are making formal software verification feasible at scale (Oct 22)

Claude autonomously converts existing fuzz tests into formal specifications and proves "no-crash" properties for 239 entry points across 123 open-source libraries, generating ~180M lines of Rocq proof code (proofs averaging ~600K lines for binaries with a median 18,000 RISC-V instructions); Anthropic calls it the first large-scale AI-driven formal verification of production software at the assembly level, and the process surfaced bugs — including a 20MB-input out-of-bounds write — that had survived years of fuzzing.

Claude Models on Google Cloud (Anthropic managed) — GA (Oct 22)

Launches a new deployment option where Anthropic directly operates the Claude stack inside a customer's Google Cloud project — auth via Google Cloud IAM, billing via Marketplace — with same-day feature parity to the native Claude Platform (including Managed Agents beta and Fast mode for Opus 5.5); models available: Fable 5.1, Sonnet 5.5, Haiku 5.5, Opus 5.5.

Open call for guest investigators in the life sciences (Oct 22)

Opens applications for life-sciences researchers to run collaborative special projects with Anthropic.

Anthropic's public S-1 still missing from EDGAR (Oct 23)

Updates yesterday's filing rumor: despite reports Anthropic could publicly file its IPO prospectus "as early as Friday Oct 23" (targeting a mid-November Nasdaq listing, after last week's disclosed $80.7B H1 net loss), no public S-1 had appeared on SEC EDGAR as of this report — the confidential draft submitted in June remains the only confirmed filing.

🔌 Claude Code Plugins

Plugin tracking skipped this run — marketplace install-count cache unavailable, per instructions.

🛠️ Skills

None of the top 5 updated in window.

Unchanged in window: addyosmani/agent-skills (last commit Oct 13), anthropics/skills (last commit Oct 9), kharmanskyi/open-steps (last commit Oct 5), K-Dense-AI/scientific-agent-skills (only an automated security-scan-report bot commit on Oct 19, no content change), cloudflare/security-audit-skill (last commit Sep 14).

🤖 Agents & Subagents

nicobailon/pi-subagents v0.81.0 (Oct 22)

Ships Windows startup-performance improvements and, in a follow-up commit (Oct 23), fixes the Pi API placement bridge.

davila7/claude-code-templates (Oct 23)

Pushes a routine automated catalog content update; no new features beyond the catalog API and stack-viewer pane already covered yesterday.

Unchanged in window: wshobson/agents (last commit Oct 17), VoltAgent/awesome-claude-code-subagents (last commit Oct 11), anthropics/commerce-agents (last commit Aug 31).

🔗 MCPs & Integrations

ChromeDevTools/chrome-devtools-mcp v1.12.0 (Oct 22)

Adds hot-reload config support and parallel task execution, fixes heap-snapshot bugs, and in a follow-up commit (Oct 23) adds a configure_webauthn tool for virtual authenticators.

upstash/context7 (Oct 23)

Ships "Admin MCP fixes from live write flows," improving library rule management, policy approval, and API response handling.

DeusData/codebase-memory-mcp (Oct 22)

Merges a fix for external package imports shadowing project files with the same name.

oraios/serena (Oct 20)

Fixes a dead Jedi language-server documentation link; no functional change.

Unchanged in window: modelcontextprotocol/servers (v2.0.0 and last commit both landed Oct 19, one day before the window).

💡 Community — Workflows & Ideas

omnara-ai/omnara (Show HN, Oct 21)

Open-sources a self-hostable "managed agent platform" positioned as an alternative to Claude Managed Agents and OpenAI's Agents API — works with any model provider, stores agent history in your own Postgres, and is free to self-host under Apache 2.0; 3.2k stars, 111 points/34 comments on HN.

rbartoli/agent-usage-guard (Oct 22)

Ships a utility that guards and rate-limits Claude Code agent usage to prevent runaway cost or infinite tool-call loops; 23 points/9 comments on HN.

Intina47/context-sync v0.5.0 (Oct 23)

Releases a tool for syncing editor context between Cursor and Claude Code sessions, pitched as "start in Cursor, finish in Claude."

📰 Quick Mentions

• FlowLens — new MCP server adding structured debug tracing/inspection tools callable from Claude Code (Oct 23).
• Anthropic reportedly builds its own source-control system — unconfirmed newsletter report that Anthropic replaced git-based infra to handle Claude's autonomous code/research loops at scale (Oct 22).
• awesome-claude-code — adds "Review Inbox" and "Molecular Animation" community resources (Oct 22).
• awesome-mcp-servers — merges 10+ new MCP server listings in a single day's curation batch (Oct 23).