What's new in 3.4.1
Workflow token budgets are now opt-in
Fixes the "budget exhausted" failures reported in #89: the workflow tool no longer lets models invent or infer a token budget. tokenBudget is a pure opt-in soft gate — runs are unlimited unless you explicitly ask for a cap (or set a standing default). The authoring skill, lifecycle guidance, and the HIGH/ULTRA effort directives all carry the same rule.
Thanks to @Whamp for the fix, and to @rwese for the report that drove it.