Top 10 hottest Claude Code ecosystem items from the past 72 hours.
v1.40.0 shipped today (May 3) with per-phase model routing, dynamic failure-tier escalation, and a --research-phase flag for the plan command. At 59.6K stars, GSD's --minimal install profile cuts cold-start system-prompt overhead by 94% (from ~12K tokens to ~700), which is the single biggest token win per install in the ecosystem right now.
npx get-shit-done-cc@latestv1.7.0 landed May 1 adding cavecrew (multi-agent cave team), stats receipts, and MCP-shrink — a new MCP input-token compressor that cuts ~46% of input tokens every session on top of the existing 65% output-token reduction. Now at 52.9K stars, caveman is one of the most-installed cost skills in the ecosystem.
curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bashA Tauri-powered desktop app that unifies Claude Code, Codex, OpenCode, OpenClaw, and Gemini CLI in one GUI — 50+ provider presets, one-click MCP and Skills sync across all agents, and a system-tray switcher. At 58.4K stars it has become the go-to desktop control panel for teams running multiple AI CLI tools side-by-side.
brew tap farion1231/ccswitch && brew install --cask cc-switchCrossed 173K stars and released v2.0.0-rc.1 in April, adding a desktop dashboard GUI, operator workflows, media tooling, and a Rust-based ECC 2.0 alpha control plane. Bundles 48 agents, 182 skills, 68 command shims, and multi-language rules — still the highest-starred Claude Code harness on GitHub.
/plugin marketplace add https://github.com/affaan-m/everything-claude-codev0.12.0 shipped April 30. Nous Research's self-improving agent creates skills from experience, searches past conversations, and builds persistent user models across sessions — it hit 131K GitHub stars in ten weeks, faster than LangChain or AutoGPT. Runs Claude via OpenRouter or direct API; includes a Claude Code SKILL.md in the repo.
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bashGarry Tan's open-sourced Claude Code setup — 23 specialist skills running as slash commands: CEO, Designer, Eng Manager, QA Lead, Security Officer, Release Engineer, and more. Surpassed 10K stars in its first 48 hours after launch and now sits at 82.7K stars and 12K forks; the design-review skill catches AI slop before it ships.
See READMENow at 71.4K stars and v12.3.8 with 244 releases. Hooks into Claude Code's session lifecycle, compresses tool usage into semantic summaries via the Agent SDK, and stores them in a local SQLite+Chroma vector DB — injecting relevant context automatically on next session start. @simonscrapes compared all Claude memory systems this week and claude-mem came out on top for zero-config simplicity.
npx claude-mem installOMO is a multi-model orchestration harness at 55K stars that transforms a single AI coding assistant into 11 parallel specialists (Sisyphus orchestrator on Claude Opus 4.7, sub-agents on Gemini/Kimi/GLM). Ships 52 lifecycle hooks, 26 tools, and a 3-tier MCP system — each task auto-routes to the cheapest capable model, making it the most cost-aware harness after caveman.
42.1K stars for an AI-powered job search system built entirely on Claude Code: 14 skill modes, a Go terminal dashboard, PDF resume generation, and Playwright-powered batch scanning of 45+ career portals. The creator used it to evaluate 740 job listings and land a Head of Applied AI role — one of the most concrete real-world Claude Code skill systems to ship this year.
git clone https://github.com/santifer/career-ops.git && cd career-ops && npm install && npx playwright install chromiumUltra-lightweight personal AI agent at 41.6K stars; v0.1.5.post3 released April 29. Keeps the core agent loop small and readable while supporting MCP, multi-platform chat (Telegram, Discord, Slack, WeChat), 24/7 market analysis, and full-stack software engineering tasks. Built as an open alternative to Claude Code's architecture — inspects Claude Code's design philosophy but runs any LLM.
pip install nanobot-ai