Asset Graph Desk

Mid-production the question is never whether a prompt is good. It is what has to be made again. Change the lead's coat and eleven of fourteen shots are now wrong, plus the character sheet all eleven are drawn against - and the sheet has to be rebuilt before the shots, or the second pass inherits the first pass's mistake. Change the style instead and it is nineteen things, because style is the one entity that reaches every shot in the project. Working that out by reading your own files is where projects quietly desynchronise, so the free in-page engine computes it by reachability over a typed graph: six entity types, five asset kinds each carrying the tier that fixes its rebuild order, and ten documented change kinds that each touch different fields and different assets. It returns every shot, asset and line that is now stale, which fields of each, and a regeneration order that never rebuilds something from a thing that is itself about to change. It also finds the dangling references, orphans and duplicate ids that would make all of that confidently wrong - a graph with a broken edge answers every later question with an undersized number, and 'only two shots are affected' is the most expensive sentence in production when the real answer is nine. Five lanes: build the graph, ask what a change costs, turn that into a batched work queue, audit the graph itself, or price three ways to take the same note. Lanes derived from the open-source MIT skill jijiutong/ai-visual-director. Not affiliated with or endorsed by jijiutong.

Every public app is built from a security-scanned skill and must pass a clean scan — skill and frontend — before it can be listed. Have a skill of your own? Turn it into an app — or read the step-by-step walkthrough.