ColoringArt

how i’m cutting token usage in Fable 5.2 while building a game 👇 my character agent had three helpers running at once and racked up 122 million cached input tokens over a long session here are the rules i’m using now: no subagents one agent, one assignment, no extra helpers quietly expanding the workload only read what the task needs our decision log is around 150 KB. loading the whole thing for every small fix adds unnecessary context. give the agent relevant files and a short brief one narrow task per session define the result, finish it, stop. don’t turn a small asset fix into an endless conversation carrying the project’s entire history keep Fable on Blender in my workflow, Fable handles the assets. Opus handles game integration, copy, and tests limit the review loop one final before-and-after sheet, one planned test pass, a two-line handoff. repeat checks when something fails or changes, not out of habit the current assignment: fix the starting gate, the emblem above it, the cart’s cargo, and the plants in the windows only the relevant Blender files. rebuild only the changed pieces. push to a separate branch and stop a clear finish line is part of the token budget