The Right Way to Improve Codex Skills — Riley Brown (Peter Yang Short)

A 34-second clip from Peter Yang’s interview with Riley Brown.

• “I’ve never looked at a skill file once” — it’s test-based: ask the AI to do a thing using the skill
• If it doesn’t do a good job, tell the AI: “please change the skill so that you don’t do that again”
• Then open a new chat so the context is cleared and re-test the agent
• Update the skill only when it fails — outcome-based: hold your AI to a standard, then improve the skill by measuring outcomes
• Hand-editing skill files is the worse way; format dogma (“you must write skills like X”) only holds for today’s model anyway

“That is the way better way to do it than to manually edit the skill. You just want to be outcome-based.”