
小巴 | Rotifer
2K posts

小巴 | Rotifer
@web3xiaoba
@rotiferdev 创始人 | Evolution Engineering 提出者 Gene = AI 能力的缺失层——可进化 · 可评估 · 可迁移 300亿 Token 实战 → 一个开源进化框架


10.3% of Vibe Coding apps have critical vulnerabilities. cURL killed its bug bounty. GitHub lets repos disable PRs. Code generation cost → zero. Quality selection → still manual. What if selection was automated? That's F(g). rotifer.dev #RotiferProtocol #AI





Rotifer v0.8.1 — Ecosystem Reach Chaos experiment: 3 APIs break, agent auto-switches genes. → 83.3% vs 33.3% uptime (2.5x) → 3 auto gene switches, 0 human intervention → New: --from-clawhub migration rotifer.dev/blog/api-apoca… #RotiferProtocol #AIAgents








Rotifer v0.7.9 — Trust Shield We scanned the Top 50 ClawHub Skills. 100% triggered at least one finding. → V(g) scanner: 7 rules, A-D grading → badge.rotifer.dev: trust badges for any README → 13 WebMCP tools with browser-native execution → MCP docs: 28/28 tools documented rotifer.dev/blog/v0.7.9-tr… #RotiferProtocol #OpenSource

.@karpathy's autoresearch: let an AI agent mutate training code, evaluate in 5 min, keep or discard, repeat. Sleep → wake up → better model. This is natural selection for ML. But what if it worked for *all* agent capabilities? 🧵



Your AI agent can now search, inspect, and rank Rotifer Genes — directly from your IDE. npx @rotifer/mcp-server Works with Cursor, Claude Desktop, and any MCP-compatible client. @rotifer/mcp-server" target="_blank" rel="nofollow noopener">npmjs.com/package/@rotif…
#RotiferProtocol #MCP

If you could pick ONE thing to fix about current AI agent tooling, what would it be? 1. Security (sandboxing, permissions) 2. Portability (cross-framework) 3. Quality (auto-selection of best tool) 4. Discovery (finding the right tool) Reply with your number. #RotiferProtocol #AIAgents


Why we replaced "Skill" with "Gene" in our AI agent framework. A thread on what the entire agent ecosystem got right — and where it stopped too early. 🧵 #AIAgents #OpenSource







