fun-game-design

Designs and verifies game fun through measured references, signed metrics, and bot playtests.

Updated Sep 21, 2026
One-click install
npx skills add https://github.com/dev-skill-up/ai-skills --skill fun-game-design-dev-skill-up
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: fun-game-design
Source: https://github.com/dev-skill-up/ai-skills/tree/main/plugins/dev-skill-up/skills/fun-game-design
Command: npx skills add https://github.com/dev-skill-up/ai-skills --skill fun-game-design-dev-skill-up

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) and assets (resource) components.

What problem does it solve? Game features often ship technically correct but feel flat because appeal is never the objective during implementation, and magnitudes typed from intuition default to timid. This Skill forces emotional targets, dial settings, and measured reference budgets before any gameplay or animation code is written, then verifies fun through deterministic checks, bot playtests, and blind judges instead of taste. ## Core Features & Use Cases - Pre-code design procedure: Commits per-component emotional targets, three dials (failure, scarcity, intensity), boundary rules for mixed designs, and a signed metric table in a design-decisions artifact before any code. - Reference measurement: Replaces guessed durations, angles, and scales with frame-by-frame measurements of a shipped game in the target genre, recorded as the project's budget. - Perceive loop and bot playtests: Renders frames, runs deterministic checks with negative fixtures, plays the loop with persona bots producing signed metrics, and uses blind judges whose disagreements fix the checks. - Use Case: When asked to make a potion-pour animation feel good, it measures a reference game first, replaces the small-angle cap, linear fill, and crispEdges quantization rather than tuning around them, then verifies the result on rendered frames. ## Quick Start Use the fun-game-design skill to review my game mechanic and produce a design-decisions document before writing any gameplay code.

Frequently Asked Questions about fun-game-design

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make a game mechanic actually fun instead of just working?

Commit to emotional targets and three dials (failure, scarcity, intensity) per component before writing code, then sign every metric from those dials. Verify the result with bot playtests, rendered-frame checks, and blind judges rather than declaring it fun from reading the source.

How to fix a game animation that feels weak or flat?

Measure a shipped reference game frame by frame before changing any duration, angle, or scale. Then replace the approximation-protecting caps (small-angle clamps, linear fills, crispEdges, rebuild-all render paths), add anticipation, overshoot-and-settle, arcs, and follow-through, and verify on rendered frames with negative fixtures.

Can this approach work for cozy games without fail states?

Yes. Cozy is a named preset with all three dials at the safe end: failure as hazard, abundance instead of scarcity, capped intensity. Metrics are re-signed accordingly, so skill gradient is measured in richness rather than speed and escalation gets a ceiling as well as a floor.

Does it work for narrative games without a repeatable loop?

Bot playtesting is skipped entirely for pure narrative games and walking sims. Beat-level pacing measurement, path coverage, and the judged-artifact protocol replace loop metrics, while the feel pipeline and reference measurement still apply unchanged.

Why not just use retention metrics to measure fun?

Retention is produced by compulsion as readily as by fun, so it is rejected as a fun proxy. In designs without a compulsion layer, voluntary quit/skip rate is the primary outcome metric, and compulsion mechanics like daily timers are named and given the sign of the scarcity dial.

What happens when automated checks pass but the judge scores low?

The check is treated as wrong, not the judge. A second independent judge with fresh context is consulted before changing anything, and where both judges agree against the checks, the deterministic checks are adjusted to encode what the frames show.