Prompt Optimizer

Problem

prompt-optimizer is for draft prompts that already state a task but lack clarity on audience, output format, constraints, or success criteria. Without these details, models may give generic answers or invent constraints the user did not want.

The goal is not to make prompts longer. It is to rewrite the same requirement into several directly usable variants so users can compare which structure fits the task.

How it works

The skill starts with two checks:

• Intent check: it restates the original request in one sentence to reduce misunderstanding.
• Clarification: if audience, output format, or boundaries are missing, it asks one or two targeted questions before optimizing.

Then it generates four independent prompts:

• A — Detail expansion: adds context, output format, examples, edge cases, and completion criteria; useful for deterministic outputs.
• B — Role-play: adds expert persona, chain-of-thought guidance, and methods such as SWOT or first principles; useful for analysis, ideation, and strategy work.
• C — Structured decomposition: splits the task into 3-7 steps with inputs, outputs, and validation checks; useful for complex multi-step workflows.
• D — Reverse refinement: infers the real goal from a vague request, puts the goal first, and adds few-shot examples plus iteration; useful for exploratory work.

The output is kept in a comparable shape so users can copy one version and use it directly.

Boundaries and notes

It should not change the user's original intent or add business requirements on the user's behalf. If the original prompt is already clear, the skill should avoid unnecessary packaging. It follows the input language: Chinese input yields Chinese prompts, English input yields English prompts. For highly specialized, regulated, or strict data-format tasks, human review is still needed for field definitions, permissions, and acceptance criteria.