game-golden-screenshot-verifierlisted

# Game Golden-Screenshot Verifier A one-command runtime verifier an agent can run unattended to answer "does the game still boot and render correctly?" The runner (`run_eval.py`, next to this file) ships in the plugin and never moves into the project; the project owns a small **seam folder** holding everything game-specific. Requires Python 3.10+ and `uv` (dependencies are declared inline in the runner and install on first `uv run`), plus a real desktop session for screen capture and input. Throughout, `<skill-dir>` means the directory containing this SKILL.md, and `<seam-dir>` means the project's seam folder. ## First-time setup in a project 1. Create the seam folder — `harness/game-golden-screenshot/` by convention (any location works; all relative paths in the config resolve against it). 2. Copy `<skill-dir>/eval_config.example.json` to `<seam-dir>/eval_config.json` and fill in the executable path, launch arguments, window title, capture size, threshold, timeout, and warmup/settle delays. Relative paths resolve against `<seam-dir>`, so an executable elsewhere in the repo needs `../` segments or an absolute path. 3. Copy `<skill-dir>/drivers/input_stub.py` to `<seam-dir>/drivers/` and replace its `drive()` body with the game's real route to a repeatable visual state (launch flag into a fixed scene, scripted menu path, or a static screen needing no input — the stub's docstring ranks the options). The stub only moves the mouse and taps Shift; it