Commentary on X
This is the cleanest cut of the stack. Thin prompt = what you want right now. Thick artifacts = the real files, the real context, the real mess. Thin Skills = when to fire and what job, not a novel in the chat box. Most people still invert it: fat prompts, empty folders, zero Skills. Dynamic loading is the point. The model should not eat the whole manual every turn. It should know the name, when to pull it, then open the Skill.
6:50 PM · Jul 15, 2026@trq212 Yall lost the mandate of heaven with your token burn and model lobotomizing. Openai eating your lunch that you originally stole from it
6:51 PM · Jul 15, 2026@trq212 Totally agree. The better way to use agent like claude code is preparing artifacts that may be needed to solve the problem
8:35 PM · Jul 15, 2026@trq212 For real, the magic is in the context, not the prompt. Less typing, better results.
7:27 PM · Jul 15, 2026View all
@MatthewBerman @EricBuess @garrytan Yeah, it depends a bit on what the skills do eg having a lot of scripts in the skills is great, using progressive disclosure is great, etc. But I think most skill.MDs are too big
7:24 PM · Jul 15, 2026ideal prompting technique is: - thin prompts - thick artifacts + context - thin skills
4:43 PM · Jul 15, 2026@trq212 @EricBuess @garrytan said thin harness fat skills I’m getting whiplash!
7:15 PM · Jul 15, 2026