A prompt you tuned for weeks, gone in the next window

You spent a week turning a review prompt into something that actually works in ChatGPT.

Tomorrow you are in Claude. The day after, Cursor. You dig through chat history, bookmarks, a graveyard doc. The leak is not “ask better this time.” It is a prompt you already tuned, typed from scratch again.

CodeExpander does not host a model. It turns a stable draft into an abbreviation. If the window can type, the abbreviation can expand.

Snippet editor

One snippet per prompt

Group by job, not by vendor: ai-review, ai-sql, ai-zh. See Groups.

Variable parts are FillIn. Do not keep review-py and review-rs as near-clones:


You are a code reviewer. Language: %fillmenu&name=language&opt=Python&opt=TypeScript&opt=Rust&opt=Go%
Scope: %fillarea&name=diff or files%

Rules
- Deterministic issues and checkable advice only
- %fillcheck&name=test gaps&default=List missing tests%

Fill once; same-name fields update together. Syntax: FillIn.

An unverified “20 magic prompts” pack will turn the library into someone else’s landing page. Three real uses earn a free-tier slot.

Boundary with “AI writing suites”

Grammar, long-form generation, scraping a CRM field into a web form — other products. This one guarantees that a prompt you already accepted does not vanish when you change chat windows.

Harvest good paragraphs from model output back into the library. That is editing, not autogeneration. The search window AI assistant can help rewrite a snippet. The authority is still the local row.

Steps: AI programming, AI use cases.

Do not

Do not put API keys in prompt bodies. Abbreviations expand in any text field.
Free tier is 50 snippets. When the prompt library grows, price or delete dead ones.

Next: Support templates are a form.