Skills aren't judgment: your digital twin needs failure lessons written in, not just templates
sujingshen · x · 2026-09-15
- Responding to the hot take that Skills are the most valuable private digital asset in the AI era, the author adds: having Skills isn't having judgment. Most people pack in portfolio pieces, workflow templates, and tone samples — a polished resume, not a brain.
- What makes a twin act like you is different: which aesthetic red lines trigger a veto, which approaches you tried and abandoned and why, which clients you take and which you refuse. If failure lessons never enter the library, the twin is just a pretty shell and you re-teach it who you are every session.
- The quoted post demonstrates a working system: WorkBuddy + Karpathy LLM Wiki, structuring half a year of work into 19 knowledge pages and a personal content Skill with open-sourced prompts across a four-layer architecture.
More from coding & agent
- In the Vibe Coding era, React is becoming a programming language — yihui_indie · 2026-09-15
- Dev builds MCP server so agents analyze his X account data before drafting posts — DarkyZ_ · 2026-09-15
- Make Claude your harshest critic: adversarial review and one-shot example workflows — thisdudelikesAI · 2026-09-15
- 10 Claude workflows the author claims can 10x your output — thisdudelikesAI · 2026-09-15
- Five exercises find four defects in the author's own LLM-judge eval system — alexpran · 2026-09-15
- Agent design tips: keep decision traces and always give users control — goyalshaliniuk · 2026-09-15