Meta prompt toolkit
A reusable meta-prompt that turns any use case into a high-quality, ready-to-use LLM prompt — Claude-first, with Gemini/GPT fallbacks.
npx -y skills add hmbseaotter/meta-prompt-toolkitAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 18 days oldThe repository was created 18 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when the user wants help writing, generating, or improving a prompt for an LLM - e.g. "write me a prompt for...", "turn this into a prompt", "help me prompt for...", "improve this prompt", or when they hand over a use case and want a polished, reusable prompt back. Produces a finished, copy-ready prompt (not advice about prompting). Also applies when the user references the CoSTAR framework or asks for a prompt built to a named structure.
SKILL.md
2.1 KB, 411 tokens by cl100k_base, as published. Nobody here has run it
Meta-Prompt Skill
Turn a USE CASE into ONE finished, copy-ready prompt, or improve an existing prompt. Deliver an actual prompt, not general advice.
How to run
- Read
core/meta-prompt-core.mdand follow its steps - it is the engine and it governs. - Apply the branch file for the model the user will RUN the prompt on - not the model you are:
- Claude (the default here) ->
branches/anthropic.md - Gemini ->
branches/gemini.md - ChatGPT / GPT ->
branches/chatgpt.mdBranches override the core where they conflict; they exist because the vendors genuinely disagree (on roles, sampling parameters, repetition, and chain-of-thought). If the user hasn't said which model, ask - it changes the output.
- Claude (the default here) ->
- If unsure what the finished output should look like, read
examples/worked-example.md. It shows one full USE CASE pass and one IMPROVE pass. - If the user is building a multi-step working session, agent, or skill - not just a one-shot
prompt - also offer the rules in
guardrails/collaboration-rules.md.
The two non-negotiables
- Never guess silently. Ask up to 3 focused questions, or state a labeled "Assumption:".
- Deliver a prompt, not advice. THE PROMPT in one code block, then WHY IT WORKS, SETTINGS, HOW TO IMPROVE.
Everything else - the four essentials, technique selection, the optional blocks - lives in the core file. Follow it there rather than from memory.
Presenting choices to the user
When a decision is needed, present it as a short list, recommended option first with a one-line reason and trade-offs, and leave room for the user to choose otherwise. Don't bury questions in prose.