PRISM refracts what you ask into its component pieces — objective, platform, constraints, style — then recombines them into an engineered prompt tuned to the exact AI you'll run it on. Then refine it in plain language until it's right.
Your request is split into labeled pieces — what's stated, what's inferred — so nothing important is left implicit.
The prompt is written in your target model's native conventions: XML for Claude, role framing for GPT, descriptor strings for image models.
Every section carries the reasoning behind it, plus which tools to run it with and why — so you learn as you go.
Tell PRISM what to change in plain words. It rewrites in place and keeps a running history of your revisions.
PRISM · prompts engineered by decomposition
authored end-to-end by Claude Sonnet 4.6 in this artifact's runtime — no keys, no backend
Describe what you want. PRISM refracts it into pieces, then recombines them into an engineered prompt for the model you choose.
One or two sentences is enough. Rough is fine — the next step surfaces anything missing.
Everything here feeds the prompt. Edit any value, remove what doesn't apply, add what's missing. Dashed cards were inferred, not stated.
Answer what you can. Anything left blank uses the suggested default and is marked as an assumption in the prompt.
Pick where you'll run this. The prompt adopts that model's conventions — and the workspace takes on its colors.