Claude Report — 2026-09-14

🚀 Anthropic Official

Claude Code v2.1.269 (Sep 11)

A feature-dense release: adds claude plugin eval to run a plugin's eval suite with a scored JSON+HTML report, makes /output-style [name] work over Remote Control and cloud sessions, attaches a diff of files changed by Bash commands to tool results, adds OTEL_METRICS_INCLUDE_REPOSITORY, and adds CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS (1–256) to raise the concurrent-subagent ceiling. Also fixes prompt-cache partial invalidation on truncated responses, F1/F2/F4/Delete handling in kitty/rxvt-unicode/WezTerm, deny/ask rules starting with ! leaking scope, CMYK JPEG attachment failures, plugin MCP servers not reconnecting in resumed remote sessions, and several VS Code extension issues.

Claude Code v2.1.270 (Sep 12)

Fixes a regression v2.1.269 itself introduced: read-only git commands in Bash started re-asking for permission after a session had been running a while. Nothing has shipped past v2.1.270 through Sep 14.

claude-agent-sdk-python remains frozen at v0.2.152 (Sep 2) — now 12 days behind the TypeScript SDK with no new package release, despite Claude Code advancing two more versions in that span.

🔌 Claude Code Plugins

Plugin tracking skipped this run — official marketplace install-count cache unavailable.

Unchanged in window: n/a (tracking skipped).

🛠️ Skills

K-Dense-AI/scientific-agent-skills

Adds a new AlphaGenome skill plus a skill-image generation script and refreshed illustrations for the AlphaGenome, Folklore, and Genomic-intelligence skills, resuming activity after a quiet day (Sep 13–14).

ayghri/i-have-adhd

Merges a Grok plugin install layout and a Persian README translation, resuming activity after stalling since Sep 10 (Sep 12–14).

Unchanged in window: tt-a1i/archify (no commits since Sep 13, already reported), affaan-m/ECC (no commits since Sep 12, already reported), agentskillexchange/skills (only automated sync-bot regeneration commits).

🤖 Agents & Subagents

nicobailon/pi-subagents

Lands workflow-script parameterization, a TUI async-widget fold-on-click, and drops the Pi 0.85.0 compatibility shim (Sep 13–14), still without a new version tag past v0.67.0.

VoltAgent/awesome-claude-code-subagents

Merges new Auth Engineer and Webhook Engineer subagents and a storage-agnostic Memory Curator subagent, resuming activity after stalling since Sep 7 (Sep 14).

Unchanged in window: anthropics/claude-agent-sdk-typescript (still v0.3.270, Sep 12, already reported), anthropics/claude-agent-sdk-python (still v0.2.152, Sep 2, see Anthropic Official), anthropics/commerce-agents (no commits since Aug 31).

🔗 MCPs & Integrations

None of the top 5 updated in window.

Unchanged in window: DeusData/codebase-memory-mcp (no commits since Sep 13, 10:10 UTC, already reported), oraios/serena (no commits since Sep 12, already reported), upstash/context7 (no commits since Sep 11, already reported), ChromeDevTools/chrome-devtools-mcp (no commits since Sep 11, already reported), github/github-mcp-server (still v1.12.1, Sep 8).

💡 Community — Workflows & Ideas

tech-leads-club/agent-skills (Sep 12, trending)

Runs Snyk static analysis in CI before publishing skills, positioning itself as a security-vetted alternative in a marketplace it says has "13% of skills containing critical vulnerabilities"; gained +265 stars on Sep 14 alone.

agent-guardrails-kit — round 12 (Sep 13)

Fixes a bug where a literal space character in a guarded path (versus a newline) silently switched every protection hook off, continuing the project's adversarial-testing series past round 11.

kruzovic7/ai-data-extractor (Sep 11)

Ships a free extractor for AI coding-assistant chat histories (Claude Code, Cursor, Windsurf, Aider, Cline/Roo Code), reaching 505 stars and 132 forks in three days.

awesome-claude-code — Claude Style Patch (Sep 14)

Adds a drop-in CLAUDE.md prose-style patch that bans habits like announcing a point before making it, colon-hinged sentences, and verbless fragment openers.

sigmastratum/sigma-operator-stack (Sep 11)

Persists project state between coding-agent sessions as append-only, explicitly-accepted records instead of reconstructing context from chat history, and re-checks an application fingerprint on each read to invalidate stale verification passes.

📰 Quick Mentions

• Dario Amodei: "We Must Pace the Frontier" — Sep 12 essay committing Anthropic to give third-party evaluators permanent employee-level access to its training pipeline; OpenAI and xAI publicly endorsed within hours.
• BlackRock Advisor Center plugin merged into claude-plugins-official — New plugin added Sep 13, outside the tracked top-10 install ranking.
• Tell HN: Claude Code appends a commit signature even with the setting off — Sep 12 bug report that a "Claude-Session" URL line gets added to commits regardless of the disabling config.
• nxm-memory MCP server — Sep 13 launch claiming 87% token reduction via structural code maps and semantic-search doc retrieval instead of full-file loads.
• Replay — audits silent prompt-cache misses — Sep 14 local tool that finds which turn broke prompt caching in agent transcripts and quantifies the re-billed token cost.
• Novgraph — persistent knowledge graph for codebases — Sep 12 tool that stores a queryable graph of commit reasoning and change impact for coding agents to query.
• awesome-mcp-servers batch update — Sep 13 merge of eight new server listings (MCP Gatehouse, ToolTrace, TokenChit, AnkusDrive, and others).
• gap-trap — Sep 13 repo-gate tool that keeps AI-generated code correct without full manual review.
• CC-Monitor — Sep 12 tool that audits every action Claude Code takes on the local machine.
• GVS5H — Sep 12 open-weight ledger-based self-orchestration technique that lifts Qwen3.8-27B past Claude Fable 5 on LiveCodeBench Hard.