You can reuse prompts across ChatGPT, Claude, Gemini, and Cursor when the prompt describes a stable job instead of depending on one tool’s conversation history.
The practical approach is to maintain one readable source, then add the current context and any small tool-specific note when you run it.
Identify the stable core
The stable core usually contains:
- the task to perform
- the inputs required
- the constraints that matter everywhere
- the expected output
- what to do when evidence is missing
For example, a prompt that reviews a product brief can require the brief, audience, constraints, and decision deadline. None of those requirements needs to belong to a particular model.
Keep changing context as inputs
Do not bake the current project, customer, or document into the reusable instructions. Use clear placeholders:
Task: Review <brief> against <decision_criteria> for <audience>.
Return:
1. Short assessment
2. Evidence for each criterion
3. Missing information
4. Recommended next revision
Do not infer missing constraints. List them as questions.
When you run the prompt, replace or accompany those inputs with the real material. The reusable source remains unchanged.
Put tool-specific notes at the edge
Sometimes a workflow needs a small adaptation, such as where you attach a source, how you reference selected code, or which output format you want from a particular surface.
Keep those notes after the core prompt:
## Run notes
- Chat tool: paste the source after the prompt.
- Coding tool: include the relevant files and tests as context.
- Long document: split the source only if the tool cannot accept it in one run.
The notes can change without fragmenting the task into four separate prompts.
Test portability with a second tool
Before calling a prompt portable, run it with the same sanitized input in two tools. You are not looking for identical wording. You are checking whether both responses satisfy the same output contract.
Use this checklist:
- every required input is explicit
- the task makes sense without the old thread
- the output can be judged by the same criteria
- model-specific advice is labeled and optional
- missing information is handled rather than invented
- no private context is embedded in the saved source
If the result fails in one tool, revise the shared instructions before creating a separate version. Split only when the tasks have genuinely diverged.
Avoid silent copies
Copying the same prompt into several notes creates a maintenance problem. One version gets fixed while the others keep the old mistake.
Choose one source file as the maintained version. If you need a tool-specific variant, name the difference in the title or description and link the variants in your own file structure.
Use a local source and a copy-based workflow
Promptlight keeps Markdown prompts in a vault folder you choose, indexes them for local search, and lets you copy a prompt into the AI tool where you want to run it. There are no direct ChatGPT, Claude, Gemini, or Cursor integrations. That is what keeps the maintained source independent from the destination.
For the ownership model, see Local-First Prompt Manager and Markdown Prompt Library. If most of your prompts start in one chat product, compare a local prompt library with chat history.
