Bookmarks
Top 10 GitHub Repos
- anthropics/skills
- Anthropic's public implementation of Agent Skills for Claude — folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized, repeatable tasks. This is the reference implementation behind the agentskills.io standard, so it's the baseline other skill packs and tools are being measured against.
- google/skills
- Agent Skills for Google products and technologies, including Google Cloud, installable via npx skills add google/skills with per-skill selection. It's under active development, but its existence confirms the skills format is going multi-vendor rather than staying Anthropic-specific.
- PrimeIntellect-ai/prime-agent
- An open-source coding and research agent built around the Recursive Language Model (RLM) abstraction, which treats context as variables ("prompt-as-a-variable") to handle long-running autonomous tasks. Pairs with PRIME-RL and pi-mono for self-improvement loops, making it one of the more research-forward coding agents trending right now.
- addyosmani/agent-skills
- Production-grade engineering skills for AI coding agents that encode the workflows, quality gates, and best practices senior engineers use — organized around DEFINE → PLAN → BUILD → VERIFY → REVIEW → SHIP phases. Practical, opinionated skill packs meant to make agents follow consistent process across every phase of development.
- NVIDIA-NeMo/Switchyard
- A Rust proxy and library for LLM traffic that routes requests across providers while preserving native OpenAI and Anthropic API compatibility. It also translates between the two API formats, records operational metrics, and supports benchmarking plus cost/performance optimization — useful as a drop-in layer for apps that want provider flexibility without rewiring their SDK calls.
- paperclipai/paperclip
- An open-source Node.js server and React UI for orchestrating a team of AI agents running business workflows — pitched as "if OpenClaw is an employee, Paperclip is the company." It's aimed at managing fleets of agents in one place rather than driving a single agent session.
- BoundaryML/baml
- A TypeScript-like programming language for agents where types persist at runtime and there's no any or unsafe casting — every feature is built to make agents make fewer mistakes. It compiles faster than Go and targets the messy boundary between structured program logic and LLM output.
- elizaOS/eliza
- An open-source TypeScript framework and product stack for autonomous AI agents — the monorepo includes the core runtime, the Eliza app, CLI, cloud services, native bridges, and first-party plugins. It's positioning itself as a bootable agentic operating system rather than a single-purpose agent.
- TencentCloud/TencentDB-Agent-Memory
- A team-level memory hub for AI agents that converts conversations, docs, and code into four reusable memory assets: Chat Memory, Skill, LLM-Wiki, and Code-Graph. The assets are governed, shared, and equipped across agents and frameworks — tackling shared context rather than per-agent isolated memory.
- pingdotgg/t3code
- An "agent harness control surface" that lets you control the agents on your machine — Claude Code, Codex, Cursor, Grok Build, and OpenCode — from a mobile app, web app, or Electron desktop app, using your own existing subscriptions. It's a remote control for local agent harnesses rather than another agent itself.
Honorable Mentions
- firecrawl/firecrawl
- The web context API for agents: search, scrape, and convert pages into clean Markdown or structured data at scale.
- browserbase/stagehand
- SDK for browser agents with self-healing actions, on top of familiar Playwright-style APIs in TS, Python, and Go.
- anomalyco/opencode
- Open-source, terminal-first AI coding agent with broad model support and a one-line install.
- corsairdev/corsair
- Unified integration layer for agents: connect once, get every integration, without the agent ever seeing credentials.
- superradcompany/microsandbox
- Fast, local-first microVM runtime for running untrusted agent workloads, user code, and CI jobs in hardware isolation.