game-prototypes β Vision & Design (Seed)¶
Seed document for this branch/repo. This repo is a monorepo of small, LLM-driven game experiments. Each prototype exists to cheaply prove or disprove a specific βis this fun?β hypothesis before any full build investment.
Vision¶
A personal lab for LLM-native games:
- LLM should do what scripted logic cannot (open-ended input, real reasoning, emergent behavior, generative content).
- Every prototype is a fast fun-test, not a polished game.
- Shelved prototypes are still successful when they produce a clear lesson.
Primary question for every prototype: Does this require an LLM and is it fun with open input?
Environment and delivery constraints¶
- Development is optimized for Claude Code on phone.
- Prototypes should run/test on phone natively where possible.
- Desktop testing can be done over Tailscale.
- All prototypes should pipe through one shared web page with a prototype switch selector for fast context switching.
Design process¶
- Use SDD + TDD in slices, scoped at human PR review level.
- Build tracer-first vertical slices that prove the feeling/mechanic.
- A prototype or major unit of work is an epic.
- Milestone/spec names are short, tag-like, and ELI5; no numbered milestone/spec ordering.
- Plan for slices to launch in agentic waves.
Documentation and governance¶
Maintain these artifacts in-repo:
- Spec docs
- Brainstorm docs
- ADR decision logs
Workflow expectations:
- One slice per PR with sober review.
- Prefer gated merges.
- Use Conventional Commits.
Suggested repo shape¶
game-prototypes/
README.md
docs/
vision-and-design-seed.md
specs/
brainstorms/
adr/
web/
index.html # shared launcher + prototype switch selector
_engine/ # thin shared LLM/state/router harness
<prototype-tag>/
README.md
spikes/
slices/
notes.md
Status tags:
exploringfun-provennot-fun-shelvedgraduatinggraduated
Seed prototype epics¶
persuade-the-guardinterrogationsecret-ruleadaptive-adversaryhidden-traitoremergent-threeliving-villagespellsmithimprov-dungeonagent-mud(graduation target, not first tracer)
Principle: Prove the feeling with the cheapest ugly tracer first, then graduate only what is truly fun.