# Game Seed Expansion > Take a fragment of a game idea — a theme, mood, mechanic, art style, setting, character, or feeling — and expand it into 3-5 wildly divergent design directions, each grounded in specific reference games. Use whenever the user shares a loose game concept, says they "want to make a game about X", has a vibe but no game yet, names a single mechanic with no game around it, or is stuck on one direction and needs alternatives. Apply this skill aggressively at the start of any game-ideation conversation, even if the user has not explicitly asked for "directions" or "options" — designers usually do not know they need divergence until they see it. - Skill: `kbravh/game-seed-expansion` (Agent Skill) - Install (CLI): `npx skillmds@latest add kbravh/game-seed-expansion` - Raw SKILL.md: https://api.skillmd.com/api/skills/kbravh/game-seed-expansion/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: kbravh (https://skillmd.com/u/kbravh) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/kbravh/game-seed-expansion --- # Game Seed Expansion The hardest moment in game design is the one where a designer has a fragment — a mood, a single mechanic, a setting — and converges on the *first* game that fits it before considering the others. Your job is to make that convergence cost something. Generate enough divergent directions that the designer has to choose deliberately, not by default. ## What a seed is A seed is anything smaller than a full game concept. Common shapes: - **A mechanic** ("a grappling hook that only works on living things") - **A mood** ("the loneliness of an empty mall at 3am") - **A setting** ("a city built on the back of a sleeping giant") - **A constraint** ("you can never see your character") - **A reference moment** ("I want it to feel like that one scene in Spirited Away") - **A character or relationship** ("an old detective and the AI that replaced him") - **A real-world activity** ("I want a game about competitive bread baking") Identify which shape the seed is. The shape changes what to do with it. ## Step 1: Find what made the seed stick Before generating directions, ask why *this* seed. Designers carry hundreds of fragments. The one they bring to a conversation has a reason. The reason often *is* the game. Ask questions like: - Where did this come from — a game, a memory, a piece of media, a real experience? - What is the smallest version of this that already feels right? - If you could only keep one quality of this idea, what is it? The answers narrow what counts as a valid direction. A "lighthouse keeper" seed sourced from *To the Moon* wants emotional resonance; the same seed sourced from *Dredge* wants slow-burn dread. The directions you generate must serve the underlying pull. ## Step 2: Generate 3-5 wildly divergent directions Spread, not optimization. The directions should: - **Span genres.** If the obvious read is a cozy sim, include a horror version, an action version, a multiplayer version, an abstract version. Force range. - **Span scope.** One should be a one-hour indie. One should be a 40-hour epic. One should be a thing that lives on a single screen. - **Span perspective.** Same seed, very different player relationships to it (first-person witness, omniscient manager, embodied participant, abstract operator). - **Include at least one "wrong" answer.** A direction that seems off-brand for the seed but reveals something the others miss. The designer often discovers what they want by recoiling from it. Each direction should be one or two sentences, not a paragraph. Vagueness here is a feature — you are mapping a possibility space, not pitching finished games. **Example (seed: "a grappling hook that only works on living things"):** 1. *Co-op horror.* You and a partner are tethered to enemies you cannot kill; the game is about coordinating retreats and using the tether to drag threats into hazards. Like *Lethal Company* if every monster was also your only mobility tool. 2. *Single-screen action puzzle.* A side-view brawler where every fight is solved by chaining grapples between enemies. Like *Smash TV* meets *Spider-Man*'s web combat, but the enemies are the level. 3. *Quiet exploration.* A wordless game where the world is overgrown and you traverse by tethering to birds, deer, and stranger things. The grapple is a relationship, not a weapon. Like *Journey* with *Worms*' rope. 4. *Roguelike about consent.* Every creature you grapple must be calmed first through a small ritual; the run is about which beings you bond with and what they let you do. The "wrong answer" direction — but it might be the most interesting one. 5. *Boss rush.* The entire game is one mechanic, applied to ten enormous creatures whose anatomy is the puzzle. Like *Shadow of the Colossus* if Wander only had the grapple. ## Step 3: For each direction, name the closest existing games Reference comparisons are not laziness — they are precision. "Like Stardew, but…" communicates more in five words than a paragraph of abstract description. For each direction: - Name 1-3 specific games that occupy the same neighborhood - Name what *this* direction would do differently from them - If you cannot find references, that is information — either the direction is unusually original, or it is unusually vague. Investigate which. Cite specifically. Not "soulslikes do this." Try "Sekiro replaced healing-as-resource with healing-as-skill-expression by tying it to posture; this direction could do the same with the grapple." ## Step 4: Find the version only this designer can make After mapping directions, ask: of these, which one does the designer have an unfair advantage in making? Their taste, obsessions, prior work, lived experience, technical skills, or community. The strongest version of a seed is usually the one the designer is *already* qualified to make, even if it is not the most obvious or commercial. Surface this directly. "Direction 3 is the one only you would make — you have been [gardening / serving in your church / running a homelab / learning Japanese] for years, and that texture would be impossible to fake." ## Antipatterns to watch for - **Generating five versions of the same direction.** "Cozy farming sim with a twist" five times is not range. If your directions could all coexist in the same genre shelf, regenerate. - **Picking the best one too early.** The designer's job is to choose. Yours is to make the choice meaningful by giving them real alternatives. - **Filtering by "what would sell."** Brainstorming is the wrong stage for market analysis. Let the wild directions stay on the list. - **Padding directions with mechanics.** Each direction is a *concept*, not a design. Save the mechanics for after the direction is chosen. - **Treating the seed as fixed.** Sometimes the right move is "this seed is fine, but the seed *behind* this seed is more interesting." Be willing to dig one layer down before expanding outward. ## When to stop Stop expanding and pick a direction when: - The designer is reacting strongly to specific options (loves, hates, "definitely not") - Two directions have started to feel like the same game in different costumes (a sign of convergence on a real idea) - You have generated five and adding a sixth feels like reaching The point is not to find the "right" direction. It is to make the designer's eventual commitment feel earned rather than defaulted into.