Game Dev Loop

Game Dev Loop — idea → docs → build → test → polish

One loop that chains the vault's game-dev skills into a single build process. It does not replace any of them — it is the sequence and the handoff between them.

game idea → GDD/TDD → engine+template → implement → test/debug → quality loop → guide → memory
  └────────────────────← bug / goal unmet / quality short ←────────────────────┘

When to Use

• Starting a real game build (web, single-file, Phaser, Three.js, or UE) that needs design, docs, code, testing, and polish chained in order.
• A task large enough that "just write code" would drift — you need a GDD and a plan on paper first.
• The user asks to go "from idea to playable", or to ship ONE polished scene.

The loop

1. Persist the goal. Capture the objective once; keep it intact for the whole run. (goal-persistence)
2. Classify the idea. Pick the archetype and the right engine/template before writing anything. A mismatched engine is the most expensive mistake. (game-archetype-classifier, or reason inline for non-2D ideas.)
3. Write GDD + TDD. Game Design Document first, then the Technical Design Document that makes each GDD section a buildable contract. (gdd-markdown-template, technical-design-document)
4. Scaffold. Pick the engine/template and get a runnable skeleton before adding features:
   • 2D Phaser → phaser-gamedev
   • single-file zero-dependency HTML → single-file-html-game
   • parallel multi-agent DDD swarm → kimi3-game-gen
   • Three.js PBR / high-fidelity 3D → three-pbr-workflow
   • AI-editable 3D engine (core/adapters + tokenized world) → cat-game-architecture
5. Implement in vertical slices. One feature → one test → repeat. Don't write all code then all tests. (test-driven-development / tdd)
6. Test + debug. Deterministic E2E with screenshots/assertions; for UE shader/GPU issues use the UE-specific skills. (playwright-testing, webapp-testing, ue4-shader-debug, ue-renderdoc-auto-capture)
7. Quality loop. Screenshot → evaluate → improve, until it meets the bar. (render-quality-loop for visuals; intro-scene-until-perfect for the single-scene polish discipline)
8. Playthrough guide. Turn the verified probes into a human-readable how-to. (guide-from-probes)
9. Write long-term memory. Decisions, gotchas, and next steps land in the durable store the same turn they happen.

Loop back: a failing test, an unmet goal, or a quality shortfall returns to the plan (step 3), not straight to more code.

2D / template-based variant: for a 2D template project, game-dev-agent-workflow is the concrete phase script for steps 2–6 (classify → scaffold → design → assets → config/registration → implement → verify). Use it instead of re-deriving those phases; game-template / game-template-engineering cover scaffolding and the KEEP/COPY hook discipline, game-debug the living debug protocol, game-build-verification the pre-ship checklist.

Pillars → skills

Rules

1. Goal first, never narrowed. Re-inject the objective every turn; audit completion against real state, never self-declared.
2. Classify before code. The engine/template choice is the most expensive decision; make it in step 2, on paper.
3. GDD before TDD before code. A game without a GDD drifts; a GDD without a TDD isn't buildable. Don't skip either for a real build.
4. Vertical slices. One feature → one test → repeat. Don't write all features then all tests.
5. Measure quality. "It runs" is not a result; a screenshot evaluation against a bar is. (render-quality-loop)
6. Write memory as you go. Decisions and gotchas land in the durable store the same turn they happen.
7. Loop back on failure. Return to the plan (GDD/TDD), not to more code.

Stage handoff

Each stage hands a written artifact to the next:

• goal → objective + status + completion evidence
• classify → archetype + engine/template choice + why
• GDD/TDD → design doc + engineering spec (each section a build contract)
• scaffold → runnable skeleton (proves the engine choice before features)
• implement → code + what changed (vertical slices)
• test/debug → red-green results + fixed error signatures
• quality → screenshot evaluation vs bar + improvements
• guide → playthrough how-to (from probe assertions)
• memory → durable facts / decisions / lessons appended to the store

Per-round gate (answer at the end of EVERY round)

A round does not close until all four questions are answered explicitly. Do not advance to the next step — or repeat — on an unanswered gate.

1. Success / fail criteria? State the acceptance bar for THIS round: what must be true (tests green? quality met? blocking issues = 0?) for the round to count as success vs failure. No criteria = drift, not a round.
2. What should touch / not touch? State the scope boundary: which files / areas / behaviors this round may modify, and which are off-limits. Anything outside the boundary is a scope violation, not a fix.
3. Report criteria? State what the end-of-round report must contain — use the fixed report format below, nothing else as the round summary.
4. Self-review rounds + next step? State how many critic rounds you will run before accepting (source-anchored-design: repeat until blocking = 0; set a hard cap if the round has a budget), and whether the next step auto-starts and repeats.

Round report (fixed format)

At the end of every round, emit exactly this block — nothing else as the round's summary. The agent does NOT judge success/failure itself; it restates the original criteria, reports each one's current status, and gives only a confidence score so the human can judge:

success criteria: <restate this round's original criteria verbatim>
criteria status: <one line per criterion: met / not met / partial, with evidence>
success confidence: <0-10>
failure confidence: <0-10>
touch: <files/areas modified>
not touch: <files/areas deliberately left alone>
next: <single next action>
self review status: <critic rounds run, blocking issues remaining>
next step status: <auto-start | wait-for-user | done>

The two confidence scores (0–10) are the agent's own estimate of how likely the round succeeded and how likely it failed — they are NOT a verdict. The human decides success/failure from the criteria status, not from the scores.

Termination

The loop ends when the goal's completion audit passes against real state — tests green, quality meets the bar, memory and guide written — and the final round report shows all criteria met with next step status: done. The agent never declares the loop a success itself; the human does, from the criteria status. A failing test, an unmet goal, or a quality shortfall is not termination; it is a return to step 3 (plan).

When NOT to use

• Trivial one-shot edits — a one-line change does not need the whole loop.
• A task one skill already covers — don't compose for the sake of composing.
• Pure asset/art generation with no code — use the asset-direction or creative skills directly.