Make prompts measurably better before sending them.
When to Use
Dispatching subagents → Run the prompt through orator_optimize first. Better prompt = less back-and-forth, more accurate results.
Writing system prompts → SKILL.md files, agent instructions, tool descriptions. Small improvements compound over many invocations.
Prompt feels vague or under-specified → Score it. Orator identifies weak dimensions and applies targeted techniques.
Quick Reference
orator_optimize(prompt: "...", intent?: "code|analysis|creative|extraction|system", techniques?: ["xml-tags", "few-shot"])
Output: Before score → techniques applied → optimized prompt → after score.
Already good? One-line confirmation: 🪶 ━━ already well-structured (8.4)
When to Skip
Skip for trivial prompts, simple questions, or prompts already scoring above 7.0. The overhead isn't worth it for single-step instructions.
Common Mistakes
| Mistake | Fix |
|---|---|
| Optimizing everything | Focus on high-leverage: subagent prompts, system prompts |
| Ignoring the score delta | Close before/after scores mean the prompt was already good |
Not using techniques override |
When you know which techniques apply, force them |