settled-reach/whatsinagame/skill/references/customization-examples.md

Customization Examples

How The Settled Reach (the project this kit was extracted from) customized the base patterns. Use these as inspiration for how far you can take the templates.

The Settled Reach: Overview

A top-down immersive sim with occlusion-based detective mechanics. Godot 4 client + Rust/bevy_ecs simulation server. Single-player with asymmetric information as core mechanic.

20 Agents with Themed Names and Personalities

Instead of generic names, The Settled Reach gave every agent a name from the project's science fiction universe. Each personality was tuned to the specific role and project context.

Name	Archetype	Unique Flavor
tyre	architect	Named after a setting element. Evaluates engine choices, client-server architecture, rendering pipelines. Specializes in Godot + Rust trade-offs.
si	project-manager	Manages sprint cycles across 6 team branches. Coordinates parallel workstreams.
hoshe	qa-engineer	Writes GDScript tests, Rust tests, and integration tests. Verifies simulation determinism.
gestalt	designer	Systems design specialist. Evaluates whether mechanics create "interesting decisions." Maps system interactions.
dudley	developer-backend	Rust simulation server specialist. Entity systems, deterministic tick processing, bevy_ecs.
stig	developer-frontend	Godot client specialist. Rendering, UI systems, fog of war, input handling.
mellanie	content-author	Writes in-game text: internal monologue, dialogue, descriptions, news ticker content.
troblum	consultant	Technical sparring partner. Evaluates architecture decisions alongside tyre.
araminta	visual-designer	Art direction, sprite standards, UI component patterns, color palettes.
inigo	audio-designer	Soundscape design, ambient layers, diegetic audio cues, spatial audio specs.
oscar	developer-backend (networking)	Client-server communication specialist. Network protocol, sync mechanisms.
qatux	librarian	Maintains decision records, discussion archives, briefings, Qdrant search index.
ozzie	stakeholder (wow-factor)	"Does this make players feel something?" Evaluates excitement and emotional resonance.
paula	stakeholder (narrative)	Political depth, conversation systems, faction mechanics, consequences.
gore	stakeholder (themes)	Philosophical questions, ascension paths, what the game is fundamentally ABOUT.
nigel	stakeholder (replayability)	Sandbox advocate. "What happens the SECOND time you play this?"
miri	worldbuilder	Setting designer. Factions, cultures, technology, locations, history, lore consistency.
justine	devops	Build pipelines, performance profiling, platform packaging, release quality.
tiger	localization	Translation infrastructure, cultural adaptation.

Key lesson: Themed names make agents feel like team members, not tools. The project's fiction naturally provided memorable, distinct names.

Game-Specific Commit Scopes

Standard scopes were replaced with project-specific ones:

agents, skills, docs, briefings, discussions, schema, db, config,
engine, simulation, client, ui, audio, assets, meta

Compare with generic: feat, fix, docs, chore, test, refactor

Key lesson: Commit scopes should match your project's actual domains. A web app might use: api, frontend, auth, db, deploy, infra, docs.

6 Worktree Branches

Instead of one main branch, the project uses worktrees for parallel development:

Branch	Worktree	Team
server	../server/	dudley, oscar
client	../client/	stig
copy	../copy/	mellanie
audio	../audio/	inigo
visual	../visual/	araminta
ci	../ci/	justine

Each sprint branch contains the full repository. Sprint briefings are written per-team (docs/sprints/sprint-N/server.md, client.md, etc.). Merges happen via PR review on main.

Key lesson: Sprint branches (sprint-N/team) with ephemeral worktrees via isolation: "worktree" give each agent an isolated copy without persistent worktree management.

Qdrant + Ollama for Semantic Document Search

The project uses local Qdrant (vector database) and Ollama (embedding model) for semantic search across hundreds of design documents, discussion rounds, and decision records.

# Search across all project documents
tooling/db/qdrant-search "asymmetric information design"

# Index a new document
tooling/db/qdrant-index docs/briefings/tyre.md

# Health check
tooling/db/qdrant-health

Configuration:

{
  "qdrant_url": "http://tower-of-joy:6333/",
  "ollama_url": "http://tower-of-joy:11434/",
  "embedding_model": "nomic-embed-text",
  "collection": "commonwealth",
  "dimensions": 768,
  "distance": "cosine"
}

Key lesson: Semantic search becomes valuable when you have 50+ documents. Below that, grep is fine. The full profile includes Qdrant; the standard profile uses grep-only /docs-search.

Custom Skills

Beyond the standard kit skills, the project added domain-specific ones:

Skill	Purpose
render-sprite	Render 3D models to 2D sprites via Godot pipeline (4 directions, 3 resolutions)
gen-audio	Generate audio assets using Stable Audio Open API
gen-image	Generate visual assets using Gemini image generation

Key lesson: The skill system is extensible. Use /skill-create to build project-specific skills for any repetitive workflow.

Domain-Specific Decision Files

Instead of a single decisions.md, the project organizes decisions by domain:

decisions/
  README.md           # Domain index with query examples
  architecture.md     # D-008, D-009, D-010, D-012, D-020, ...
  perception.md       # D-011, D-015, D-016, D-017, ...
  content.md          # D-023, D-024, D-025, ...
  scope.md            # D-001, D-003, D-005, ...
  process.md          # D-004, D-021, D-022
  questions.md        # Q-001 through Q-011
  rejected.md         # R-001 through R-010

Each decision has an ID, status, date, and summary. The README provides an index and grep-friendly query examples.

Key lesson: Start with 2-3 domain files (architecture, scope, process) and add more as needed. The standard profile deploys these three. The full profile adds domain files based on the project type.

Workshop Orchestration

Multi-agent design workshops with structured rounds:

1. Workshop brief written to docs/workshops/workshop-N/brief.md
2. /workshop-start parses the brief, spawns participants as teammates
3. Rounds proceed with facilitator, participants discuss, decisions are captured
4. Output: decision records, updated domain files, next-steps

Workshops are used for major design decisions (game mechanics, architecture choices, scope planning). Smaller decisions go through async discussion rounds.

Key lesson: Workshops are the full profile's killer feature. They turn "Claude discussing with itself" into structured, multi-perspective design sessions with real output.