CloakBrowser vs SkillClaw

CloakBrowser is an open-source stealth Chromium browser that defeats bot detection by patching fingerprints at the C++ source level — not through JavaScript injection or flag tricks that break on every update. With 49 C++ patches covering canvas, WebGL, audio, fonts, GPU reporting, screen properties, and WebRTC, it achieves 0.9 reCAPTCHA v3 scores (human-level) and passes Cloudflare Turnstile, FingerprintJS, and 30+ other detection systems out of the box. It's a drop-in replacement for Playwright and Puppeteer — swap one import line and your existing automation scripts work with zero other changes. An optional humanize=True flag adds Bézier-curve mouse movements, character-by-character typing, and realistic scroll patterns for behavioral detection evasion. Native SOCKS5/HTTP proxy support with GeoIP-matched locale makes multi-geo scraping seamless. With 7,800+ GitHub stars and 1,600+ gained today alone, it's clearly scratching a massive itch. The source-level patching approach means it survives Chrome version updates — a longstanding pain point that killed previous tools like undetected-chromedriver. It's fully open source, free to use, and auto-downloads its binary on first pip/npm install.

SkillClaw is a research framework from Alibaba's AMAP-ML team that enables collective skill evolution for LLM agent systems deployed at scale. The core idea: instead of each user's agent interactions existing in isolation, SkillClaw aggregates anonymized skill-improvement signals across all users to continuously refine a shared library of reusable agent skills — without requiring centralized fine-tuning. The framework introduces a three-component architecture: a Skill Extractor that identifies and catalogs atomic capabilities from interactions, a Skill Evolver that proposes improvements based on aggregate feedback, and a Skill Selector that routes tasks to the best-available skill version per user context. Published on April 9 and hitting #1 on Hugging Face trending papers this week with 277 upvotes, the paper reports significant improvements over per-user baselines on complex multi-step agentic tasks. This matters especially for production agent deployments where cold-start problems are severe — a new user's agent immediately benefits from millions of prior interactions. It's a fundamentally different model of agent improvement than either fine-tuning (expensive, periodic) or RAG (retrieval-only, no learning).