A prompt answers once. A skill stacks prompts and can clock in every day.

Somewhere in your chat history sits the best prompt you ever wrote. You know it worked: the output came back sharp, structured, exactly what you needed. It is also gone, buried under six months of conversations, and next time you need it you will either scroll and hunt or type a worse version from memory.

This is the default fate of good instructions. Used once, admired once, lost. Nothing accumulates, so every session pays full price again for something you already figured out.

A skill is the fix: take the instruction that worked and save it as a file the assistant can find and load on its own. Give it a name, a job description, a procedure, an output format. The next time a request matches, the model reads the file and follows it. Same model, better equipment.

The escalation path has three rungs. A prompt is a request you type. A saved prompt is a request you paste. A skill is a request promoted into infrastructure, sitting where the assistant can reach it, working whether you remember it or not.

Two things change when instructions become skills. Consistency: the task runs to the same standard every time, because the standard is written down instead of remembered. And compounding: every improvement to the file upgrades every future run of the task, permanently. Fix a flaw once and it stays fixed.

Skills also stack. One skill can call others, check their work, and route a large job to the right specialist in the right order. That routing layer, an orchestrator, is how a single command like “take this brief and give me a presentation” becomes possible.

The prompt library below is raw material. The rule for promoting anything in it is simple: use it twice, and it deserves to be a skill.

Go Deeper