optimize-promptlisted
Install: claude install-skill woznet/wozdev
# Optimize Prompt
Rewrite or draft a prompt so the model reading it gets everything it needs and nothing else.
**Core rule:** every token must change what the reader does. Keep the task, the situation, the constraints, the reasons behind non-obvious rules, and the output format. Cut what repeats, flatters, hedges, or tells the model what it would do anyway. Shorter is the result, not the goal: a prompt that loses a needed fact costs more than the tokens it saved.
## Quick Reference
| Topic | Do | Don't |
| --- | --- | --- |
| Opening | State the task and the situation in the first lines | Greetings, "You are a helpful assistant", announcing that this is a prompt |
| Instructions | One statement per behavior, placed where it applies | The same rule in the intro, the body, and a "remember" footer |
| Reasons | A short reason beside any rule that is not self-evident ("read aloud by text-to-speech, so no ellipses"); it covers the cases the rule did not list | Bare commands, or a paragraph of justification |
| Wording | Plain full sentences addressed to the reader ("Return the ID") | "I would like you to please make sure that you..." |
| Positive form | Say what to do ("Answer in prose") | Only what to avoid ("No bullets"), unless the prohibition is the point |
| Emphasis | A calm statement, once, with its reason | Capitals, "CRITICAL", "!!!", repeated warnings; they make the reader apply the rule too broadly and rigidly |
| Specifics | Numbers, names, formats ("under 100 word