Paul Bakaus 分享高级智能体技能工程与形容词级设计控制

!!!!

!!!! Paul Bakaus @pbakaus most agent skills are a system prompt and a prayer. they produce safe, median output because that's what the model defaults to. next week at the @aiDotEngineer world's fair i'm giving two talks on how to break past that ceiling: 🛠️ the dark arts of skill engineering (1h workshop) mon 6/29, 4:30pm · track 2, room 2006 after building 24+ skills in one uber skill across 9 harness/model combos (see @impeccable_ai and agent-reviews), i, um, acquired some obscure, but powerful knowledge, stuff like: → spawn parallel sub-agents that independently judge the same work, then force their conflicting takes into one result → mixture-of-expert skills that route to specialized sub-agents, the way frontier models route to specialized networks → skill memory via persistent context files that restore across sessions, so every run builds on the last → hooks that auto-fire after a skill to validate, transform, or chain into the next one → barely-documented env vars and shell expansion to make skills context-aware before they even run 🎨 design at the speed of adjectives (design engineering track) wed 7/1, 1:30pm · room 2014 design tools today sits at the wrong altitude. padding sliders and color pickers are too low. prompt-to-design tools that one-shot a landing page are too high, and more dangerous, because it looks like it's working. no design director hears a prompt and starts pushing pixels. the brief comes first. the control that's actually missing: adjectives. "make it feel warmer." "add tension." "strip it to its essence." the things a creative director actually thinks in. drawing on @impeccable_ai , my open-source design tool with 24 adjective-level commands like /bolder, /quieter, /distill. come argue with me at either one. 🔗 View Quoted Tweet 💬 0 🔄 1 ❤️ 6 👀 1609 📊 1 ⚡