sean goedecke

LLM-generated "skills" (explanatory prompts for specific tasks) work better when created after solving a problem rather than before. A recent paper found that pre-generated skills provide no benefit because they bake in incorrect assumptions from training data. The effective approach is to have the LLM solve the problem through iteration first, then distill that learned experience into a reusable skill document. This captures knowledge gained from millions of tokens of problem-solving rather than just regurgitating existing training data.

LLM-generated skills work, if you generate them afterwards

Interesting take. The idea that skills are more useful as distilled learning after solving the problem, rather than as a starting artifact, really landed for me

Beyond being more effective, skill creation post-solve just feels right.
I was iterating on worktree management workflows, had Claude Code work through opening new terminal panes per worktree, then told it to turn that into a skill after the fact. So quick, and I never had to explain the skill!
What’s also underrated is how easy it is to refine the skill afterward. I noticed the panes should be named after their worktrees so cleanup stays reliable even if you move them around. One quick instruction and the skill was updated.​​​​​​​​​​​​​​​​