Claude: Ecosystem & Tools

• Growing ecosystem of GitHub repos: Superbase CLI, Skill Creator, Get Shit Done, Notebooklm-py, Obsidian.md — curated list of repos that extend Claude Code capabilities. (from claude code github repos)

• Skill Creator: Notable tool for building Claude Code skills programmatically. (from claude code github repos)

• Obsidian integration is a recognized community pattern: Obsidian + Claude Code is repeatedly mentioned as the go-to second brain stack. (from claude code github repos)

• Demand for productivity/execution frameworks: The "Get Shit Done" repo signals that people want structured execution workflows around Claude Code, not just chat. (from claude code github repos)

• gstack is MIT-licensed and open source at github.com/garrytan/gstack, freely forkable for Claude Code skill development (from gstack open source garrytan)

• Garry Tan (YC president) personally uses Claude Code skills as his primary development workflow, signaling mainstream adoption among top-tier founders (from gstack open source garrytan)

• SkyPilot agent skill enables Claude Code to autonomously manage GPU clusters — provisioning 16 GPUs on Kubernetes, submitting jobs, checking logs (from parallel gpu autoresearch skypilot)

• Skills that teach coding agents to manage infrastructure are a new paradigm — SkyPilot skill lets Claude handle GPU provisioning end-to-end (from parallel gpu autoresearch skypilot)

• shadcn/ui has official Claude Code skill at ui.shadcn.com/docs/skills — works out of the box, Claude picks it up and correctly uses shadcn components (from shadcn skill claude code)

• @trq212 maintains a pinned thread cataloging Claude Code technical writing, cross-posted to official Claude blog — blog becoming destination for in-depth Claude Code guides from X (from thariq technical writing thread)

• Community-curated resource repos for Claude Code and Codex (docs, workflows, video tutorials) are forming, confirming a maturing practitioner community (from claude code codex resource list)

• The Claude skill-creator now includes built-in test generation to measure and optimize skill trigger rate — the rate at which a skill is correctly invoked when it should be (from claude skill creator test generation)

• Skills are portable across three Claude surfaces: Claude Code (plugin), desktop app, and Cowork — meaning skill investments aren't locked to one product (from claude skill creator test generation)

• shadcn/cli v4 adds presets, dry-run mode, and monorepo support alongside "shadcn/skills," giving teams safer automation in CI/agent-driven pipelines (from shadcn cli v4 skills)

• "Agent councils" — groups of agents reviewing PRs and app performance — represent a new organizational pattern where AI agents collaborate on quality assurance (from chintan agent first engineering)

• Building a "deep library of agents + skills" as reusable components is the agent-era equivalent of internal libraries — the compounding asset is now prompt/skill infrastructure (from chintan agent first engineering)

• Agent-first teams generate internal tooling at dramatically higher rates because the cost of building a tool drops to near-zero — 30+ tools in weeks (from chintan agent first engineering)

• The "underrated tools" question for Claude Code generated 188 replies (45% reply-to-like ratio), indicating a deep feature surface with non-obvious power-user techniques ripe for community knowledge sharing (from underrated claude code tools)

• Garry Tan publicly released gstack as an installable skill pack (6.4K likes), demonstrating massive demand for pre-configured Claude Code setups — skill packs are becoming a "dotfiles for AI" culture (from garry tan gstack skill pack)

• The cf-crawl skill uses Cloudflare's /crawl endpoint to batch-crawl documentation sites in one shot, outputting markdown; installable via npx claude-code-templates@latest --skill utilities/cf-crawl, establishing a package-manager pattern for skill distribution (from cf crawl scheduled knowledge base)

• Intercom built an internal plugin system with 13 plugins, 100+ skills, and hooks, demonstrating enterprise-scale investment in Claude Code as a "full-stack engineering platform" (from intercom claude code plugin system)

• 100+ skills across 13 plugins at a single company suggests skill libraries grow rapidly once teams adopt the plugin pattern, reinforcing the need for skill organization and management tooling (from intercom claude code plugin system)

• Karpathy's autoresearch tool can be repurposed to iteratively tune AI agent skills by running automated research loops in the background — combined with Hamel Husain's evals-skills framework, it creates "auto-evals" for skill quality (from autoresearch skill tuning evals)

• Tuning 190+ skills requires automated pipelines rather than manual prompt engineering — running autoresearch "all day in the background" as a continuous skill improvement loop is an emerging pattern (from autoresearch skill tuning evals)

• Josh Pigford released an open-source macOS app for organizing and editing AI agent skills, confirming that managing large SKILL.md collections is a recognized workflow problem warranting purpose-built GUI tooling (from shpigford macos markdown skills apps)

• The emergence of standalone skill organizer apps mirrors how ecosystems form around developer platforms — rapid open-source tooling filling gaps the platform doesn't address (from shpigford macos markdown skills apps)

• Role-specific loop patterns expand Claude Code beyond engineering: recruiters monitoring open roles, teachers flagging struggling students, real estate agents watching MLS for niche listings (from claude code loop business use cases)

• Anthropic reached $14B annualized revenue run rate ($0 to $14B in 3 years); Claude Code alone is at $2.5B run rate in under one year, with $1M+ customers 40x'd year-over-year to 500+ (from anthropic 14b revenue run rate)

• Claude Code now accounts for 4% of all GitHub commits, indicating meaningful penetration of the global developer workflow (from anthropic 14b revenue run rate)

• Claude Code + Obsidian is described as "insanely powerful" and uniquely enabled by Obsidian's local markdown files — Notion and Apple Notes cannot replicate this because they lack direct filesystem access (from claude code obsidian power)

• A curated list of daily-use Claude Code plugins generated 2.3K likes and 207 retweets, indicating strong demand for plugin discovery and curation in the ecosystem (from best claude code plugins)

• Claude Code has unreleased features discovered via binary inspection: "Penguin Mode" (fast mode), "Grove" (privacy/data), "Amber Flint" (agent teams), "Marble Anvil" (thinking edits) — agent teams being built as a first-party feature (from claude code unreleased features)

• A comprehensive guide to building skills for coding agents (619 likes) suggests skill-authoring patterns are mature enough for canonical documentation, with architectures converging across agent platforms (from building coding agent skills)

• X's revamped API enabled a Claude Code skill for real-time social search, modeled after the native research agent pattern — skills are expanding Claude Code beyond development into live intelligence monitoring (from claude code x search skill)

• A self-learning Claude Code skill that observes usage patterns and automatically creates new skills achieves meta-programming where the agent's capability set grows through use (2.5K likes) (from self learning claude code skills)

• CodexBar tracks token usage in Claude Code, addressing the need for real-time AI spend monitoring; 1.3K likes signals cost management is a widespread pain point (from codexbar token tracking)

• Anthropic open-sourced 11 domain-specific plugins (sales, finance, legal, data, marketing, support), lowering adoption friction for enterprise teams to fork and customize (from anthropic open source plugins)

• Anthropic released a free official course "Claude Code in Action" covering best practices, recommended even for advanced users; 1.3K likes signals strong demand for structured learning resources (from claude code free course)

• Decode embeds a browser and whiteboard directly into Claude Code, creating a visual feedback loop where developers annotate UX issues while the agent codes — whiteboard-style annotation is an emerging UX pattern for human-AI collaboration (from decode browser whiteboard claude code)

• Power users invest 1,200+ hours into Claude-based research workflows across AI papers, market analysis, and competitive intelligence, indicating Claude is becoming a primary research tool (from claude research assistant prompts)

• dev-browser (npm i -g dev-browser) lets agents control browsers by writing sandboxed Playwright scripts in QuickJS WASM — fundamentally different from screenshot-based computer use (from dev browser sawyerhood)

• Pre-approve dev-browser in Claude Code settings with "Bash(dev-browser *)" permission to eliminate approval prompts — pattern for any trusted CLI tool (from dev browser sawyerhood)

• iMessage channel plugin reads chat.db directly via Bun's SQLite driver and sends via AppleScript — zero external dependencies, macOS only (from claude sms messaging setup)

• Claude Code --channels flag enables persistent bidirectional communication channels (polling inbound + sending outbound), fundamentally different from regular MCP servers (from claude sms messaging setup)

• Claude Cowork Dispatch: text Claude from your phone, it completes the task — no setup, no plugins (from claude sms messaging setup)

• Cross-platform SMS requires Twilio SMS MCP server (YiyangLi/sms-mcp-server) with a Twilio API account — the only option that works on Android/web (from claude sms messaging setup)

• iPhone Shortcuts can route messages to Claude API and send responses back — zero-code, no-server approach for phone-to-Claude communication (from claude sms messaging setup)

• Companies building agent-economy primitives: agentmail (AI email), tryagentphone (AI phone), daytonaio/e2b (AI computers), browserbase/browser_use/hyperbrowser (AI browsing), firecrawl (crawling), mem0ai (memory), composio (SaaS tools), elevenlabs/vapi_ai (voice) (from an economy of ai coworkers)

• Linear MCP expanding from engineering to cross-functional workflows — product management capabilities beyond issue tracking (from linear mcp product management)

• Tom Dörr built a senior design architect agent for Claude that handles complex design architecture tasks — specialized professional agents are expanding Claude's capabilities into domain-specific workflows (from claude senior design architect agent)

• Claude Code is being used for code auditing of Google and Meta ad systems, demonstrating practical application beyond typical coding assistance into security and quality assurance workflows (from claude code audits google meta ads)

• Evo is an open-source plugin for Claude Code and Codex that turns a codebase into an autonomous research loop — it auto-discovers metrics to measure, instruments benchmarks, and runs tree-search algorithms with parallel subagents to optimize performance, eliminating manual benchmark creation (from evo claude autoresearch orchestrator)

• A GitHub plugin marketplace for Claude Teams auto-installs multiple plugins (with sub-skills and agents) across every team member's instance, where repository updates sync automatically to all instances — centralized, single-source plugin management with no manual intervention (from claude teams github plugin marketplace)

• Tolaria ships an out-of-the-box MCP server so Claude and other AI tools can natively read and edit the knowledge vault without external integrations — a reference pattern for knowledge tools shipping MCP as a default surface; built with AI-assisted engineering (100K+ LOC, 3,000+ tests at 85% coverage, 9.9/10 code health) on Tauri/React/Rust (from tolaria llm wiki app karpathy)

Anthropic Platform Releases

• Claude Managed Agents (public beta on the Claude Platform) pairs an agent harness tuned for performance with production infrastructure — prototype-to-launch in days; positions Anthropic as the PaaS layer for agents (from claude managed agents production platform)

• Claude Managed Agents = Platform-as-a-Service for AI: 4 user personas, live console with sessions/analytics/per-agent costs, $2.58 fulfillment cost on $1k of service delivery (~99.7% margin); a free Google Doc deploys your first agent when given to Claude Code (from claude managed agents breakdown economics)

• Managed Agents API integrates with Linear's Agents SDK — Claude one-shotted a complete deployment example (linear/claude-managed-agents-demo); the platform is built to slot into existing developer-tools ecosystems, not replace them (from claude managed agents linear sdk integration)

• The Advisor Strategy on the Claude Platform: pair Opus as advisor with Sonnet/Haiku as executor for near-Opus intelligence at a fraction of the cost — splitting reasoning from execution as a first-class platform pattern (from claude advisor strategy platform)

• The Monitor tool (Claude) creates background scripts that wake the agent only when needed — major token saver, replaces polling loops, follows logs for errors and PR state changes (from claude monitor tool background scripts)

• Anthropic shipped a 14-minute video guide ("Agents → Structure → Tools → Execution → Systems → Money") and a 33-page Claude Skills guide — first-party documentation now covers persistent instructions, memory, automation, and monetization (from anthropic agent architecture guide)

• Claude Design (Anthropic Labs) is in research preview on Pro/Max/Team/Enterprise — make prototypes/slides/one-pagers by talking to Claude; powered by Opus 4.7 vision and natively renders HyperFrames videos in 2 prompts (from claude design hyperframes video creation)

• Anthropic's official claude-code-setup plugin scans a project and recommends hooks, skills, MCP servers, subagents, and automations, then walks through setup step-by-step — first-party configuration tooling is now part of the platform surface (from claude code setup plugin enhancement)

Free Inference and Model Access

• NVIDIA hosts ~80 AI models via free APIs (MiniMax M2.7, GLM 5.1, Kimi 2.5, DeepSeek 3.2, GPT-OSS-120B, Sarvam-M, etc.) — set base_url to integrate.api.nvidia.com/v1 with an API key from build.nvidia.com; plugs into OpenClaude, OpenCode, Zed, Hermes agent, and Cursor (from nvidia free ai models apis)

Skill Distribution and Community Tooling

• ZenSpinner replaces Claude Code's spinner verbs with zen quotes — small, but the trend of customizing every surface of the agent experience is itself a signal of ecosystem maturity (from zenspinner claude zen quotes)

• FieldTheory CLI (npm install -g fieldtheory, ft sync) downloads X bookmarks locally so agents can read them; ft viz, ft classify <url> — local bookmark graph as agent-readable context (from fieldtheory x bookmarks cli tool)

• Allie K. Miller's /ss screenshot skill is now widely shared as a reference implementation for visual input — /ss N <action> lists the newest N screenshots and acts on them (huh to explain, fix to debug, do this to remix) (from claude screenshot skill visual processing)

• Bilgin Ibryam shipped a Claude Code skill that auto-generates and updates Excalidraw architecture diagrams — visual documentation as workflow side-effect, not a separate task (from claude excalidraw architecture diagrams)

• ByteRover unifies scattered knowledge from Obsidian, GBrain, wikis, and forgotten markdown files into a Claude/Claude Desktop-queryable, relevance-scored index — brv connectors install "Claude Desktop" (from byterover unified knowledge search)

• Cabinet runs on Claude Code with agents that have heartbeats and scheduled jobs — open-source Paperclip-style document processor + KB + inline web app, all stored as markdown on disk with no vendor lock-in (from cabinet llm knowledge base tool)

• Lamina Labs' SDK gives Codex the ability to render full whiteboard animated explainer videos in seconds — single API call, "to draw" becoming an agent primitive (from codex whiteboard animation sdk)

• HKUDS/Vibe-Trading ships an MCP plugin so Claude/Cursor/Codex can run trading flows directly from the editor — finance-as-skill via MCP is a reference pattern for vertical agent toolkits (from vibe trading open source trading agent)

• Claude Ads audits Google/Meta/YouTube/TikTok/LinkedIn ads across 250+ checks and reportedly cuts ad waste up to 35% — packaged Claude Code skills for marketing operations are crossing into production (from ai automation github repositories passive income)

• claude-smart is an open-source local plugin for Claude Code and Codex that turns session corrections into reusable learnings across projects, reducing unnecessary planning loops and token usage on repeated task patterns (from claude smart self improving plugin)

• LibreChat's self-hosted model wrapper includes Claude alongside ChatGPT, Gemini, DeepSeek, and 20+ other providers with native MCP support, showing Claude ecosystem access increasingly happens through owned multi-model infrastructure as well as first-party clients (from free github repos replacing paid tools)

Boris Cherny Dogfooding Opus 4.7

• Boris Cherny (Anthropic) reports being incredibly productive dogfooding Opus 4.7 over weeks — first-party productivity tips on 4.7 are now part of the public discourse (from opus 4 7 productivity tips)