game-concept

1. Setting — where and when. Concrete: "1890s Appalachian coal town, after a mine fire" beats "dark fantasy."
2. Tone — e.g., dread, whimsical, melancholic, absurd, grim-earnest. One or two words.
3. Protagonist — who the player is (or plays as). Role, not name.
4. Antagonist forces — what resists the player. Can be a faction, a nature, an economy, an internal state.
5. Themes — the 2–3 ideas the game is about. Not mechanics — meaning. "Isolation," "debt," "forgiveness."
6. Sensory palette — dominant colors, key sounds, signature prop. Short. This informs art direction later.

Then capture references: 3 games + 3 non-game (film, book, place, music, painting, myth). References are stronger signal than abstract adjectives.

Worldview directly constrains genre. A tone of "dread + isolation + collapse" plus a protagonist with no combat skills rules out bullet-hell; it points at puzzle, narrative, or survival.

Phase 2 — Player fantasy & verbs

Ask: "What does the player actually do, second to second?" List 3–7 verbs. These are the honest spine of the genre.

• Draw, play, remove, upgrade, draft → deckbuilder
• Move, bump-attack, descend, identify, die → roguelike
• Jump, dash, wall-slide, pogo, respawn → platformer
• Place, route, upgrade, wait, balance → simulation / tycoon
• Aim, dodge, reload, bank, graze → bullet-hell / shmup
• Push, slide, undo, solve → puzzle
• Select, cover, overwatch, flank, end turn → tactics / SRPG

If the verbs don't cluster, the concept isn't a game yet — push the user back to Phase 1.

Phase 3 — Audience & scope

Three one-liners:

1. Scope — jam (≤ 1 week), prototype (1–2 months), commercial (6+ months). Drives what can be promised.
2. Session length — 5 min, 30 min, 2 hour+. Drives save design and content budget.
3. Platform — desktop browser only, mobile touch, both. Drives input and render decisions.

Phase 4 — Genre selection matrix

Map the verb cluster from Phase 2 to a candidate genre, cross-check against tone from Phase 1. Pilot genre skills currently available in this repo:

More genres will be added (platformer, tactics, metroidvania, bullet-hell, puzzle, simulation). If the verb cluster doesn't fit a listed genre, record it anyway as a free-form Chosen Genre and proceed to game-scaffold, which falls back to its own stack matrix.

If the user is genuinely torn between two genres: pick the one whose pillars (read in the genre skill's opening section) line up with the worldview's tone and themes. Mechanics are interchangeable; meaning is not.

Phase 5 — Write docs/concept.md

Create docs/concept.md with exactly these sections:

# <working title>

## Pitch
<one-sentence logline from Phase 0>

## Worldview
- **Setting:** …
- **Tone:** …
- **Protagonist:** …
- **Antagonist forces:** …
- **Themes:** …
- **Sensory palette:** …
- **References:** 3 games + 3 non-game

## Verbs
- …
- …
- …

## Audience & scope
- **Scope:** jam | prototype | commercial
- **Session length:** …
- **Platform:** …

## Chosen Genre
<genre name — exactly matches a game-<genre> skill name if one exists>

## Why this genre
<2–4 sentences connecting verbs and worldview to the genre pick>

The Chosen Genre line is a machine contract, not prose. Downstream skills grep for it. Keep it to a single lowercase word matching an existing genre skill (roguelike, deckbuilder, …) when possible.

Anti-patterns

• Jumping to tech first. "Should I use Phaser or PixiJS?" is meaningless before the verbs are listed. Redirect.
• Genre before worldview. "I want to make a roguelike" is fine as a seed, but still answer Phases 1–2 or the game will be generic.
• Worldview-only pitches. "Melancholy lighthouse keeper" is a short story, not a game. Phase 2 verbs are non-negotiable.
• Listing ten references. Three games + three non-game. More signals indecision.

Hand-off

• Genre now named → load game-<genre> (e.g. game-roguelike, game-deckbuilder).
• Then game-scaffold for tech stack + folder layout (the genre skill will inject its recommended stack).
• Then game-loop → game-systems → game-perf as normal.