I've been spending way too much time iterating on prompts lately. You write something, get a mediocre output, tweak a word, try again, restructure the whole thing, try again...
The biggest time sink for me isn't even the writing it's figuring out the right structure. Where does the context go? Should I add constraints before or after examples? Am I giving the AI too much, or not enough?
Instead of writing a wall of text, you decompose your prompt into visual blocks — role, context, objective, constraints, examples, output format. Then compile to the format your model processes best: XML for Claude, Markdown for ChatGPT and Gemini.
Available as:
→ Web app (no account required)
→ Chrome extension (sidebar inside ChatGPT, Claude, and Gemini)
→ Claude Code MCP (for terminal workflows)
You can also paste a rough idea and let AI break it into blocks automatically.
