gat-verifylisted

# gat-verify — prove the slice works before claiming it Request: $ARGUMENTS GAT's honesty gate. A milestone is not "done" because the code compiles — it is done when the reference is traceable, the engine lifecycle closes, the requested interaction runs, the numbers hold, the assets match the style, and every requirement is covered. Report with evidence; if something failed, say so with the output. ## Six checks ### 1. Reference and provenance Run `tools/reference_gate.py` on the slice map. Confirm cited local files exist, upstream commits and licenses are pinned, adaptations target the actual architecture, and required notices remain. ### 2. Godot runtime lifecycle Confirm pre-edit MCP preflight exists. After the final edit, require a fresh save-snapshot receipt and `run → debug output → stop` evidence. Relevant parser/runtime errors must be zero. A headless logic smoke is useful supplementary evidence, not a substitute. ### 3. Interaction and visual behavior For UI/input/navigation, exercise entrance → action → state change → exit/return using provider runtime tools or a targeted clickthrough harness. Capture screenshots/frames for layout or render claims. State exactly which method ran; startup alone is not interaction evidence. ### 4. Balance / data simulation (for systems with numbers) Do not tune by feel. Use the two-headed method (from `skill-design`): - **Static audit** — compute a power/cost index for each skill/item/foe from `data/*.json`. - **Combat/econo