godogen（godogen 搜索）

Bevy Game Generator

Generate and update Bevy games from natural language.

Capabilities

Read each stage file from ${GODOGEN_SKILL_DIR}/ only when you reach that stage.

Pipeline

User request
    |
    +- Check if PLAN.md exists (resume check)
    |   +- If yes: read PLAN.md, STRUCTURE.md, MEMORY.md, ASSETS.md if present -> skip to task execution
    |   +- If no: continue with fresh pipeline below
    |
    +- Generate visual target -> reference.png + ASSETS.md (art direction only)
    +- Analyze risks + define verification criteria -> PLAN.md
    +- Design architecture -> STRUCTURE.md + Cargo.toml + src/ stubs
    |
    +- If budget provided (and no asset tables in ASSETS.md):
    |   +- Plan and generate assets -> ASSETS.md + updated PLAN.md with asset assignments
    |
    +- Show user a concise plan summary (risk tasks if any, main build scope)
    |
    +- Execute (see Execution below)
    |
    +- If final presentation media is required:
    |   +- Read capture.md, produce a fresh screenshots/result/{N}/ bundle with raw frames and video.mp4
    |
    +- Summary of completed game

Assets

If a budget is provided, generating proper assets is part of the task, not optional. Do not fall back to procedural primitives when the budget allows a real asset. Plan and generate the asset through asset-planner.md / asset-gen.md.

Procedural stand-ins are acceptable only for genuinely abstract shapes (platforms, blockers, particles) or when the asset planner has explicitly ruled an asset out on budget grounds.

Placeholder primitives in gameplay code are a signal that the asset step was skipped. Go back and generate the asset before continuing.

Execution

Read task-execution.md before starting. Two phases:

1. Risk tasks (if any) — implement each in isolation, verify, commit
2. Main build — implement everything else, verify, present results, commit

If PLAN.md calls for presentation media, finish through the Bevy capture flow in capture.md and leave a fresh screenshots/result/{N}/ proof bundle behind.

Bevy Help

When you need Bevy-specific help, use the bevy-help skill with a targeted query. It keeps large reference material out of the main pipeline while giving you version-aware access to rustdoc, the checked-out Bevy repo, official examples, and Learn docs.

Use the skill inline when you already know what class, trait, system pattern, or example to inspect and can answer by reading a small number of docs. Use a dedicated helper agent when you need to discover candidate APIs, compare several patterns, or read multiple or large docs and reduce them to a compact answer.

Be specific about what you need:

• Targeted query — ask for a concrete symbol or pattern: "Camera3d: what is the Bevy 0.18 pattern for rendering to an image target?"
• Full API — only when you need to survey the whole type: "full API for AssetServer"

Context Hygiene

Keep important state in files so the pipeline can resume cleanly after long threads or compaction:

• PLAN.md — task statuses and verification criteria
• STRUCTURE.md — architecture reference
• MEMORY.md — discoveries, quirks, workarounds, what worked or failed
• ASSETS.md — asset manifest with paths and generation details

After completing each task: update PLAN.md, write discoveries to MEMORY.md, and commit. If the thread becomes noisy, summarize the important state into those files and continue from the artifacts instead of relying on conversational memory.

Visual Verification

Do not trust code alone — verify on screenshots, captured frames, and video. Code that looks correct often still ships broken placement, wrong scale, clipped geometry, missing elements, or bad motion timing.

When code and media disagree, trust the media. Be skeptical: the job is to find what is still broken, not to argue that it is probably fine. If a requirement is not clearly visible, treat it as not done.

Inspect captures directly while you work, then finish with a fresh screenshots/result/{N}/ proof bundle containing video.mp4 and the raw frameXXX.png sequence used to encode it.