| name | generating-retro-arcade-concepts |
| description | Batch-generates, evaluates, and specs multiple fixed-screen arcade game concepts in the style of 1978-1983 cabinets. Use when asked to brainstorm several new arcade games at once, produce a concept slate to choose from, or turn era/hardware constraints into ranked concepts plus implementation specs. For hardening a single already-chosen concept, use designing-mini-games instead. |
Use this skill to invent a slate of new small arcade games and pick the best ones, not to recreate existing classics. The deliverable is several concepts → an evaluation → a shortlist → implementation specs. If the user already has one concept and wants its mechanics hardened, that is a different job (single-design), not this one.
Hard Constraints
Every generated concept must satisfy all of these:
- Self-contained on a single screen; scrolling stages are prohibited.
- No hand-crafted complex level design; no maze navigation, platforming, or long enemy spawn tables.
- Player controls are movement plus one button.
- At most 3 types of enemies or obstacles; each enemy's behavior rule is expressible in 1-2 sentences.
- Difficulty escalates only through numerical changes (speed, count, spawn interval, hit points, score multipliers).
- A clear threat must emerge within 30 seconds.
- Randomness is allowed but must never cause unfair instant deaths.
- Lives system: a mistake consumes a life rather than ending the game immediately.
- No definitive ending; sustain infinite waves/rounds through numerical escalation.
- In attract mode with no input, the game must still produce visually interesting motion.
- Expressible within 256x224 logical resolution, 16x16 sprites, 64-color / 8x8 tile equivalents (the era default; a project brief may override these numbers).
- The audio identity must be describable in a single phrase covering its signature SFX and jingle direction.
Target And Avoid Directions
Pull toward (seeds, not templates): Game & Watch situational judgment, Space Invaders approach pressure, Centipede splitting/transformation, Frogger lane dodging, Missile Command defense decisions, Snake self-imposed constraint, Breakout reflection/destruction, Pengo push/crush/herd.
Steer away from designs where level design is the primary fun: Pac-Man maze AI, Galaga formation staging, Xevious scrolling placement, Donkey Kong platform/ladder stages, Dig Dug combined terrain+enemy pressure.
Slate Diversity Policy
Slates tend to collapse toward one economy type per run (whichever pattern is freshest in context). Diversity is enforced intrinsically via the Mechanism Signature taxonomy below — NOT by comparing against this project's own previously produced games. Do not scan the working repository's shipped-game directories (in this repo, tmp/games/<slug>/ and docs/<slug>/) as a differentiation reference. Referencing, differentiating from, and drawing on well-known commercial games remains allowed and expected (seeds above, field 10 in the format).
Workflow
- Before any ideation, shuffle the reward-conversion values from the Mechanism Signature taxonomy in
references/concept-generation-format.md and pre-assign one to each of the 5 concept slots without replacement — this makes a single-economy slate structurally impossible regardless of what was built recently. If the project brief itself constrains the economy type (e.g. "five defense games"), the brief wins: skip the axis-1 pre-assignment, note the override, and apply every reward-conversion distinctness requirement below to the risk-shape axis instead.
- Generate 5 concepts, each in the 14-field format from
references/concept-generation-format.md, honoring each slot's assigned reward-conversion value.
- Evaluate all 5 against the criteria in that reference (implementability, rule clarity, single-screen fit, low level-design dependency, risk/reward strength, replayability, differentiation, attract appeal).
- Select the 2 most promising concepts. They must differ on the reward-conversion axis; if the two best scorers share it, promote the next-best concept with a different value and say so.
- Write an implementation specification for each selected concept (screen layout, object list, state transitions, collision detection, difficulty-escalation formula, game-over condition).
Read references/concept-generation-format.md for the per-concept field list, the evaluation rubric, and the implementation-spec template.
Validation Before Finishing
- Each concept passes every Hard Constraint above (check explicitly; reject concepts that need scrolling, a second button, hand-built levels, or more than 3 enemy types).
- Difficulty escalation is stated as a numerical formula, not as new content.
- The attract-mode highlight is a concrete moment, not "it looks cool."
- Idle and button-mashing are not the dominant strategy in any selected concept.
- The slate honors the pre-assigned reward-conversion slots (no two concepts share a value), and every pair of concepts differs on at least 2 of the 4 mechanism-signature axes.
- The 2 selected concepts differ on the reward-conversion axis.
- Every concept's differentiation score went through the Nearest-Precedent Check in the reference: nearest well-known commercial precedent named with its 4-axis signature, matching-axes score cap applied, recall confidence stated.
- Each selected concept's spec is concrete enough to start implementing from.
