Game Feel

Make the intended player action work, communicate its consequence, and verify it in the running game. Menus, moving pixels, and passing unit tests can each exist while the requested gameplay is missing. Begin with a concrete play scenario rather than a generic visual score.

Establish the interaction

Read the user's latest brief and existing game contract. Record the core action, input, consequence, uncertainty or resistance, and expected feedback. Use templates/VERB_SHEET.md when that record does not already exist. Keep small changes small; an input fix does not require rewriting the GDD.

Preserve the genre and intended controls. Driving may require continuous steering and physics; a card game may use discrete selection and stay still between turns. Widgets, text, menus, and accessible alternative controls are legitimate when they serve that interaction. Do not add idle animation, drag gestures, or a new engine to satisfy a probe.

Read references/rules.md when diagnosing the shape of play or ordering a prototype. For substantial art or repeated refinement, also read references/production-checkpoints.md: functional, visual, motion, performance, and human acceptance require different evidence.

Rescue an existing game

1. Reproduce the reported problem in the actual play state. Use the project's supported runtime tools or authorized computer-use tools. Read references/verification.md for scenario design, optional diagnostic probes, and measurement limits.
2. Name the failed experience and point to its evidence: the car rotates but cannot drive, a card commits before the player can inspect its cost, or a hit has no readable consequence. A DOM count or screenshot difference alone is not a finding.
3. Trace the failure to input, state transition, simulation, camera, feedback, or presentation. Fix the earliest broken dependency in one coherent slice. Avoid redesigning working controls to match a template.
4. For feedback work, use references/juice.md. Compare the same action before and after, including release/cancel, failure/recovery, rapid repetition, and accessibility modes where affected.
5. Re-run the affected scenario and required project checks. Report observed behavior, evidence, and remaining gaps; distinguish measured response from subjective enjoyment.