Key points
- 12-factor-agents (humanlayer, 20,797 stars): Twelve engineering principles for production-grade LLM software. Core thesis: most "AI Agents" are deterministic code with LLM calls at key nodes. Agents need natural-language-to-tool-call translation, owned context windows, structured tool outputs, unified execution/business state, simple start/pause/resume APIs, human-in-the-loop escalation, owned control flow, error compression into context, small focused Agents, multi-surface triggers, and stateless reducer design. Principles 8 (own your control flow) and 10 (small, focused Agents) are the most common production pitfalls.
- CLI-Anything (HKUDS, 36,941 stars, +1,049 daily): One-command CLI generation that makes any software Agent-Native. From HKU Data Science Lab. Install via
pip install cli-anything-hub, thencli-hub install <name>. Supports Obsidian (48 unit tests, 7 e2e tests), Zotero, Inkscape, Kdenlive, Unreal Engine, n8n, Exa search, Dify, QGIS, UniMol, Safari via safari-mcp, and DomShell with URL validation and DOM sanitization. Replaces brittle UI scraping with structured CLI channels. - agent-skills (tech-leads-club, 4,161 stars, +1,244 daily): Verified, secure skill registry for professional AI coding Agents. Compatible with Antigravity, Claude Code, Cursor, and Copilot. Functions like an npm/pip ecosystem but with built-in trust gating—skills are vetted before Agents can load them.
- CloakBrowser (CloakHQ, 15,569 stars, +1,420 daily): Stealth Chromium with source-level fingerprint patching, Playwright-compatible APIs, and 30/30 pass rate against anti-bot tests including Cloudflare, reCAPTCHA, and Turnstile. Notable technical achievement but carries service-term risks; bypass tools violate most platform ToS.
- ai-agents-for-beginners (microsoft, 63,649 stars): 12-lesson curriculum covering Agent architecture, tool use, memory, planning, multi-Agent collaboration, evaluation, and safety. Signals that Agent development has entered mainstream technical training.
- academic-research-skills (Imbad0202, 12,515 stars, +1,439 daily): Academic research Agent skill pack.
- New to Agents: microsoft/ai-agents-for-beginners
- Production development: humanlayer/12-factor-agents
- Agent-accessible software: HKUDS/CLI-Anything
- Secure skill management: tech-leads-club/agent-skills
Three signals for Agent-Native development
1. From "what can it do" to "how to do it right" — 12-factor-agents marks the transition from exploration to engineering rigor. 2. Interfaces matter more than capabilities — CLI-Anything and CloakBrowser both favor structured access channels over visual recognition and click simulation. 3. Security is a first-class concern — agent-skills' 1,244 daily stars prove developers want bounded, verified capability expansion rather than unrestricted Agent freedom.
Getting started
References
1. https://github.com/humanlayer/12-factor-agents 2. https://github.com/HKUDS/CLI-Anything 3. https://github.com/tech-leads-club/agent-skills 4. https://github.com/CloakHQ/CloakBrowser 5. https://github.com/microsoft/ai-agents-for-beginners 6. https://github.com/Imbad0202/academic-research-skills 7. https://github.com/trending