| name | prompt-creator |
| description | Design custom developer prompts, .cursorrules files, AI editor configurations, developer briefs, specs, and PRDs. Use when the user asks to create system prompts, agent instructions, coding rules, or reusable prompt templates with {{placeholders}}. Outputs a single polished markdown artifact. |
| license | MIT |
Prompt Creator
When to Use This Skill
Apply when the user asks to create:
- System / role prompts for an LLM or agent
.cursorrules / AI editor configuration files- Developer briefs, specs, or PRDs
- Reusable prompt templates with clearly marked
{{placeholders}} for variables
- Coding standards / style guides for AI assistants
Related skills: For building live React/HTML artifacts use artifact-creator. For authoring Agent Skills (multi-file SKILL.md bundles) use skill-creator.
Output Contract
The interface extracts the FIRST markdown code block from your reply and loads it into an editable artifact panel. So:
- Explain your key choices briefly in the chat FIRST (2–4 sentences, outside the code block).
- Then put the ENTIRE finished artifact inside ONE markdown code block (
``` or ```markdown). Everything inside that block IS the deliverable — no commentary inside it.
- Output exactly one code block, fully complete, with no placeholder TODOs.
Default Structure
Default to markdown structure (headings, bullet lists, numbered steps) so the result is scannable and easy to edit. A well-structured prompt typically includes:
# [Role Name]
## When to Use This Prompt
[Trigger conditions]
## Core Instructions
[The main directives — what to do, what to avoid]
## Output Format
[How the response should be structured]
## Constraints
[Boundaries — what NOT to do, edge cases]
## Examples
[1–2 concrete examples of good output]
.cursorrules Format
For .cursorrules files, use a more compact directive style:
You are an expert [role].
Always:
- [rule 1]
- [rule 2]
Never:
- [anti-rule 1]
When [situation]:
- [specific instruction]
Template Variables
For reusable templates, use double-brace placeholders that users can find-and-replace:
Write a {{component_type}} called {{name}} that {{purpose}}.
Use {{framework}} with {{styling_approach}}.
Quality Bar
- High-signal, low-fluff: every line should earn its place. Cut filler words.
- Unambiguous: a model reading the prompt should know exactly what to do with no guesswork.
- Scoped: clear boundaries on what's in/out of scope.
- Tested mentally: walk through "if I gave this to an LLM, would it produce what the user wants?"
- Complete: no "TODO: add more rules here" — ship the finished prompt.
