game-design

name	game-design
description	Experienced game designer review of narrative arc, difficulty curve, character development, and mechanics coherence. Use this when asked to review game design, pacing, progression, story, or balance.

Game Design Review Process

Review the game's overall design for narrative coherence, difficulty progression, character utilization, and mechanical balance. The game is a snow groomer simulation across 11 levels at a fictional Savoie ski resort — it should feel like a well-paced learning journey from rookie to expert.

This skill is advisory only. It produces findings and recommendations but always asks the user for confirmation before making any changes. Game design decisions are inherently subjective.

Phase 1: Data extraction

Use explore agents to gather the complete game design picture from these sources:

1. src/config/levels.ts — All 9 level definitions: difficulty, dimensions, time limits, coverage targets, obstacles, steep zones, winch anchors, access paths, hazards, wildlife, bonus objectives, intro dialogue keys
2. src/config/gameConfig.ts — BALANCE constants: stamina/fuel/food rates, movement physics, slope/tumble/winch thresholds, wildlife distances, avalanche risk values
3. src/config/localization.ts (FR block only) — All dialogue keys: jeanPierreIntro, levelIntro, marieWelcome, taunt categories, tutorial steps
4. docs/GAMEPLAY.md — Player-facing level guide, character descriptions, controls, food items
5. docs/ROADMAP.md — Planned features that would affect design (special missions, advanced mechanics)

Build a level progression matrix in SQL:

CREATE TABLE IF NOT EXISTS design_review (
    id TEXT PRIMARY KEY,
    severity TEXT NOT NULL,
    category TEXT NOT NULL,
    title TEXT NOT NULL,
    description TEXT,
    affected_levels TEXT,
    status TEXT DEFAULT 'open'
);

CREATE TABLE IF NOT EXISTS level_matrix (
    level_id INTEGER PRIMARY KEY,
    name TEXT,
    difficulty TEXT,
    time_limit INTEGER,
    target_coverage INTEGER,
    new_mechanic TEXT,
    character_speaker TEXT,
    obstacles TEXT,
    has_winch BOOLEAN,
    is_night BOOLEAN,
    weather TEXT,
    bonus_objectives TEXT,
    notes TEXT
);

Phase 2: Review dimensions

Evaluate across 5 dimensions, launching parallel explore agents:

1. Narrative arc & story coherence

• Does the story have a clear three-act structure? Act I (learning the ropes, levels 0-2), Act II (rising challenge, levels 3-5), Act III (mastery & high stakes, levels 6-8)
• Do character appearances match the narrative moment? Jean-Pierre for mentoring, Thierry for hazard warnings, Marie for warmth/respite, Émilie for peer encouragement
• Is there emotional progression? Encouragement → confidence → tension → triumph
• Do failure taunts match the level's emotional tone? Early levels should be gentler, late levels can be more dramatic
• Are there missed story opportunities? Moments where a character should speak but doesn't, or narrative gaps between levels

2. Difficulty curve & mechanic introduction

• One new concept per level rule — Does each level teach exactly one new mechanic?
  • L0 (Tutorial): Basic movement + grooming
  • L1 (Green): Independent grooming, obstacles
  • L2 (Blue): Steeper slopes, tighter time
  • L3 (Park): Precision grooming, special features
  • L4 (Red): Fuel management + service roads + cat tracks
  • L5 (Park): Competition pressure, halfpipe precision
  • L6 (Black): Night operations + winch
  • L7 (Black): Avalanche hazards
  • L8 (Red): Storm weather, endurance
• Are coverage targets and time limits calibrated? Plot target_coverage vs time_limit vs level_area — is there a smooth curve or sudden jumps?
• Is the difficulty gap between adjacent levels appropriate? No level should feel trivially easy after the previous, nor impossibly hard
• Are bonus objectives achievable on first attempt? They should reward skillful play, not require foreknowledge

3. Character development & utilization

• Character frequency — Count how many levels each character appears in via dialogue. Are any underused?
• Character consistency — Does each character maintain their established voice across all appearances?
• Character purpose — Each character should serve a narrative function:
  • Jean-Pierre: progression gating, mentorship, technical knowledge
  • Thierry: safety exposition, hazard mechanics introduction
  • Marie: emotional anchor, food/rest mechanics, resort atmosphere
  • Émilie: peer motivation, competition context, modern perspective
• Character arc — Does the player's relationship with each character evolve? (Jean-Pierre: skeptic → respectful, Émilie: colleague → friendly rival)

4. Mechanics coherence & balance

• Do BALANCE values create the intended experience?
  • Stamina: Does LOW_STAMINA_THRESHOLD (30%) create meaningful pressure without frustration?
  • Fuel: Do consumption rates create urgency on fuel-management levels without being punishing on earlier ones?
  • Slopes: Does TUMBLE_SLOPE (40°) feel fair? Is there enough warning before tumbling?
  • Winch: Does WINCH_FORCE (0.3) make winch feel necessary on steep slopes?
  • Avalanche: Does RISK_PER_FRAME (0.015) create tension without being random?
• Service point placement — Are fuel stations and Chez Marie placed at narratively and mechanically appropriate locations?
• Bonus objectives — Do they align with the level's teaching goal? (e.g., fuel_efficiency on the fuel-management level)

5. Pacing & flow

• Session length — Total playtime estimate for all 9 levels. Is it appropriate for a browser game (30-60 min ideal)?
• Breather moments — Are there levels that provide respite between high-intensity ones? (Park levels should be fun/creative, not stressful)
• Level variety — Do levels feel distinct in both mechanics and atmosphere? (Day/night, weather, terrain shape)
• Replayability hooks — Do bonus objectives and time-based scoring encourage replaying levels?
• Flow state — Are time limits generous enough to allow exploration but tight enough to create urgency?

Phase 3: Cross-model consultation

For comprehensive design reviews:

• When running as Opus: consult gpt-5.2 (creative/narrative strength) and gemini-3-pro-preview (analytical/balance)
• When running as any other model: consult claude-opus-4.6

Prompt reviewers with the level progression matrix and ask them to evaluate:

• Narrative pacing and emotional arc
• Difficulty curve smoothness (plot it mentally)
• Any "dead zones" where nothing new is introduced
• Missing player motivations or unclear goals
• Whether the game delivers on its premise (learning to be a ski groomer)

Skip cross-model for minor reviews (single level tweaks, number adjustments).

Phase 4: Prioritized findings

Severity	Criteria	Examples
HIGH	Broken progression, impossible/trivial levels, missing core mechanic introduction	Level teaches nothing new, coverage target unreachable in time limit, character contradiction
MEDIUM	Suboptimal pacing, weak narrative beats, miscalibrated values	Difficulty spike too steep, character underused, bonus objective misaligned with level theme
LOW	Polish opportunities, enhanced immersion	Additional dialogue for atmosphere, better breather pacing, richer lore connections

Track in SQL:

INSERT INTO design_review (id, severity, category, title, description, affected_levels)
VALUES ('example-id', 'MEDIUM', 'pacing', 'Title', 'Description', '3,4');

Phase 5: Recommendations

For each finding, provide:

1. What — The specific issue
2. Why it matters — Impact on player experience
3. Suggested fix — Concrete change (new dialogue, adjusted BALANCE value, reordered level, added mechanic)
4. Affected files — Which source files need changes
5. Risk — Whether the fix could break existing mechanics or tests

IMPORTANT: Always ask the user for confirmation before implementing any changes. Present findings as recommendations and wait for approval. Game design decisions are subjective — the user has final say on narrative direction, difficulty tuning, and character voice.

Implementation priority:

• HIGH findings: fix in localization.ts, levels.ts, or gameConfig.ts as appropriate
• MEDIUM findings: propose changes for user approval
• LOW findings: document in ROADMAP.md for future work

Verification after changes:

1. npx tsc --noEmit — Type-check passes
2. ./run-tests.sh --browser chromium — No regressions
3. Cross-reference GAMEPLAY.md — Update if level descriptions changed

Design principles

Core experience pillars

1. Progressive mastery — Each level should make the player feel they've learned something real about snow grooming
2. Authentic Savoie atmosphere — The resort should feel like a real place with real people
3. Respectful challenge — Difficulty should come from skill, not from obscurity or randomness
4. Simulation depth — Mechanics should reflect real groomer operations (winch, fuel, weather, slopes)

Level design guidelines

• Teach, test, reward — Introduce a concept (dialogue), let the player practice (early level), then test mastery (late level/bonus)
• Safe failure — Early failures should feel recoverable and educational, not punishing
• Environmental storytelling — Level geography, weather, and time of day should reinforce the narrative
• Bonus as depth — Bonus objectives reveal the level's hidden mastery dimension without blocking progression

Balance philosophy

• Generous-then-tight — Start levels with comfortable margins, tighten on replays via bonus objectives
• Visible consequences — Players should see/feel the impact of their choices (fuel gauge dropping, snow quality visible)
• Fair warnings — Never punish the player for something they weren't told about (Thierry's warnings, tutorial hints)
• Satisfying arcs — Each level should have a mini-arc: setup → tension → resolution