game-production

Coordinate an authorized game-production task from its saved GameGen brief through a playable slice, asset reviews and integration. Use for multi-stage game work or resuming it.

SKILL.md

GitHub

Coordinate game production

Read shared context and the ready project preferences through scripts/prefs.py read. Use repository contracts for paths, ownership and asset records. If setup is incomplete, route to game-bootstrap once.

Translate game.goal into the player's objective and game.scope into this delivery's boundaries. Preserve accepted art direction and gameplay decisions. A new instruction steers the current task unless it explicitly replaces it. Inspect current files and evidence before treating inherited work as complete.

Select only the needed workflows

Work Skill
Visual direction and construction views game-concept-art
3D sources and exports blender-game-assets
Pixel, painted, cutout or rendered sprites game-sprites
Terrain tiles and adjacency rules game-tilesets
Skeletal, cutout or frame animation game-animation
Gameplay and persistence godot-gameplay
Levels, camera, lighting and 2.5D presentation game-world-building
Interface and cross-device input game-ui-input
Sound, music and action effects game-audio-vfx
Procedural music and SFX (local audio route) game-audio-procedural, through game-audio-vfx
Gameplay and visual verification game-qa
Measured runtime optimization game-optimization
Packaged delivery game-export
Trailers, gameplay videos and GIFs game-trailer

Load only the selected skill and its relevant references. Provider skills are dependencies for the selected service, not replacements for the game brief or GameGen preferences.

Build a representative slice

Choose a small end-to-end experience that proves the core loop: an actor or controllable object, one environment or screen, one meaningful interaction/challenge, feedback and completion/recovery. Set observable success criteria from the brief. Use a small asset review scene to validate the art and animation before generating the whole cast or environment library.

Follow the configured concept and asset review policy. Prepare a concrete image/model/engine capture for user review; under agent review compare against the approved references and record the selected revision. Resolve silhouette and camera readability before expensive detail. Keep rejected revisions only when useful for recovery, outside the engine import tree.

Parallel work and resumption

Delegate only when authorized by the current task and available tools, within workflow.parallel_agents and the host limit. Use concrete independent assignments with exclusive file ownership. Serialize preferences, shared scenes, asset manifests, spend reservations and live-editor mutations. Give workers the same preferences revision and an explicit import/controller contract. Reconcile contract changes before integration.

Each handoff includes readiness, changed files, generated IDs/metadata, verification and blockers. Use the existing tracker or a compact state file, with one scoped task when sufficient. Do not turn ordinary implementation into a documentation exercise. Maintain job IDs and reservations across turns so a retry or context reset does not duplicate expensive work.

Integrate and verify the slice before scaling content. Finish with actual playable behavior and evidence appropriate to the delivery scope. Make the distinction between working code, visual interpretation, measured performance and target-platform readiness explicit.