Forge Prompt

Turn a short, unclear ask into the prompt it deserved — one question at a time, in the language the user picks, ending in a block they paste into a fresh session.

Where this sits: nowhere in a chain — independent of frame-change, interpret-session, and every lane. It produces one artifact and stops; the receiving session decides for itself what the work is and how to open it, by design, not by omission. It also runs happily beside /interpret-session in a parallel window — hand that window the finished prompt and nothing else, no interview trail, no reasoning, so it reads the artifact cold.

Any subject. Code is the common case, not the boundary — a migration, a document, a dataset, a vendor conversation, a decision to be made elsewhere can all be forged.

The Iron Law

NAME WHAT IT TOUCHES. NEVER PRESCRIBE WHAT TO DO ABOUT IT.

Two halves, two different failures — and a prompt can carry both at once. Miss the first and the receiving session guesses the object: wrong-target action rises from 9.6% to 75.1% as target identity blurs, and 55.8–67.8% of acted runs cross a boundary nobody wrote down. Miss the second and it stops looking: handing a downstream agent your reasoning and your chosen direction helps up to a threshold, then converges it prematurely on your assumptions instead of on its own reading of the territory.