CommunityArte y diseñogithub.com

3dMVP/blobit-cli-skill-pack

Codex skills for Blobit CLI game asset workflows

Compatible con~Claude CodeCodex CLI~Cursor
npx skills add 3dMVP/blobit-cli-skill-pack

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentación

Blobit CLI

Use blobit as the asset backend, and as the scene/publish backend when the game should consume a hosted Blobit scene.

  • Use blobit to:
    • create projects
    • upload images
    • generate or reuse 3D assets
    • inspect placement metadata
    • compose and validate scenes
    • publish manifests
  • For browser games, use normal code tools to plan, scaffold, implement, run, and visually verify the playable game.
  • In game projects, Blobit owns runtime asset generation. Gameplay code, UI, input, camera, physics, and game state stay in the game codebase.
  • For browser games, published manifest mode is the default asset delivery path: use the manifest_url returned by blobit publish create and load it with @blobit/sdk.

When to use

Use this skill when the user asks to:

  • create or manage Blobit projects and assets
  • generate 3D models from images
  • lay out a scene using semantic placement commands
  • publish a Blobit scene for browser consumption
  • integrate published Blobit content into a web game with the SDK
  • create or update a browser game where Blobit should generate or supply the runtime assets

Rules

  • Prefer blobit --json when command output must be parsed.
  • Prefer semantic scene commands over hand-editing raw transforms.
  • For new assets, prefer Blobit image generation, image upload, or local-image upload plus model/world generation unless the user explicitly provides an existing model to reuse.
  • For model and world generation commands, prefer the stored default provider unless the task needs a one-off override via --provider.
  • Prefer blobit config set-provider <provider> over telling the user to edit Blobit config files by hand.
  • Do not use the legacy provider alias --endpoint-type in new commands; use --provider.
  • Inspect placement metadata before arranging or framing when an asset's origin, footprint, or floor contact is unclear.
  • Run scene analyze before and after major layout changes.
  • Run scene validate before publish.
  • When a task changes visible scene geometry and a local preview or viewer is available, capture a screenshot and inspect for obvious visual issues.
  • For browser games, avoid downloading assets and load the published Blobit manifest at runtime.
  • Use blobit assets download only when the user explicitly asks for bundled, offline, static, or local-file assets.
  • Do not ask for or report a dollar asset budget. The current Blobit CLI does not return asset-generation spend; ask for asset scope, asset count, and priority instead.
  • For high-polish games, use review-gated mode by default: ask brief questions, generate a visual target, write PLAN.md, generate or assign Blobit assets, build a first playable, capture proof, then iterate on feedback.
  • Before planning game implementation, ask or record the architecture choice for collision/physics: custom arcade collision, Babylon/Havok physics, or a hybrid.
  • Do not use non-Blobit asset-generation vendors or scripts when the task asks for Blobit-made assets.
  • Treat Blobit as the asset-and-scene backend only; use the repo's normal code tools to build game logic and rendering.
  • For published runtime content, use the manifest_url returned from blobit publish create and load it with @blobit/sdk.

What It Produces

  • a Blobit project_id
  • uploaded source image asset ids or reused source asset ids
  • generated model asset ids and, when applicable, completed job ids
  • a persisted scene with placed models, camera state, and semantic layout operations applied
  • scene analyze and scene validate results
  • a published manifest_url plus related publish metadata such as publish_id and viewer_url
  • the runtime integration point in the web app where @blobit/sdk loads the published manifest
  • for game tasks, PLAN.md, STRUCTURE.md, ASSETS.md, and visual proof from the running browser game when the project workflow supports them

Pipeline

  1. Inspect or create the Blobit project.
  2. Upload a source image and generate a model, or reuse an explicit existing asset.
  3. Inspect the generated or reused asset, especially placement metadata, before scene layout decisions.
  4. Add the model to the scene.
  5. Apply semantic placement operations such as drop-to-floor, place-near, arrange, align, avoid-overlap, distribute, snap, and frame-camera.
  6. Run scene analyze, review overlaps, bounds, and warnings, and adjust the scene if needed.
  7. Run scene validate before publish and treat validation warnings as a blocking issue unless the user explicitly accepts them.
  8. Publish and capture the manifest_url and any related viewer or publish metadata.
  9. In the game client, load that manifest through @blobit/sdk.
  10. Build or update the actual game logic and rendering with the normal project stack.

For playable browser games, read references/game-workflow.md before planning or coding. It defines a planning and implementation loop adapted for Blobit-only assets. For the Godogen-derived rationale behind that game workflow, read references/godogen-patterns.md only when updating the skill or reconciling workflow decisions.

Runtime Validation

  • After major layout or camera changes, run the local game, preview, or Blobit viewer when available.
  • Capture a screenshot when rendered geometry matters and inspect for floating models, clipping, wrong scale, bad orientation, bad camera framing, and overlaps that semantic checks may miss.
  • Treat screenshot review as a supplement to scene analyze and scene validate, not a replacement.
  • If no runtime preview is available, rely on CLI validation and explicitly note that visual validation was skipped.

Scene Notes

  • Use drop-to-floor when an object should visibly rest on the stage or ground plane.
  • Use frame-camera after adding, removing, or rearranging models, or when the current camera may no longer frame the subject.
  • Inspect asset placement metadata before arrange or place-near when the object's origin, support surface, or footprint is unclear.
  • Prefer semantic scene operations over direct transform edits; use direct transforms only for precise final adjustments or explicit user requests.
  • Run scene analyze after bulk arrangement or overlap-resolution passes to confirm the scene matches expectations.

Expected Outputs / Handoff

  • the key ids and URLs: project_id, relevant asset ids, job ids, publish_id, manifest_url, and viewer_url when applicable
  • a short summary of the scene operations applied and the resulting validation state
  • the exact runtime integration point where the app loads the published manifest through @blobit/sdk
  • a note about whether runtime screenshot validation was performed and what it found
  • any unresolved layout, validation, or runtime limitations that still need follow-up

Core commands

  • blobit --json projects list
  • blobit --json projects create --name ...
  • blobit --json assets generate-image ...
  • blobit --json assets upload-image ...
  • blobit --json jobs model-from-image ... --wait
  • blobit --json jobs model-from-prompt ... --wait
  • blobit --json jobs world-from-prompt ... --wait
  • blobit --json assets inspect <asset_id>
  • blobit --json assets download <asset_id> --out ...
  • blobit --json scene add-model <project_id> <asset_id>
  • blobit --json scene add-world <project_id> <asset_id>
  • blobit --json scene analyze <project_id>
  • blobit --json scene place-near ...
  • blobit --json scene arrange ...
  • blobit --json scene validate <project_id>
  • blobit --json publish create <project_id> --allowed-origin ...

Read references/workflows.md for end-to-end command recipes. Read references/game-workflow.md for browser game planning, implementation, asset manifest, and visual proof guidance.

Skills relacionados

zzorp/tesla-lightshow-skill

This is a Claude Code SKILL.md that can be used to build Tesla Light shows by analyzing the sound file.

community

DoYitNow/image-to-svg

Codex skill for converting images into SVG

community

figma/figma-create-design-system-rules

Generates custom design system rules for the user's codebase. Use when user says "create design system rules", "generate rules for my project", "set up design rules", "customize design system guidelines", or wants to establish project-specific conventions for Figma-to-code workflows. Requires Figma MCP server connection.

community

ObaidGits/kria-ai

KRIA is a Rust-powered desktop AI assistant with a Python sidecar for ML processing, featuring local LLM orchestration, voice pipelines, web/search tools, document RAG, image generation, system automation, and safety-audited tool execution.

community

OpenSenseNova/SenseNova-Skills

Modular SenseNova skills for building AI-powered office assistants and productivity workflows

community

homeassistant-ai/home-assistant-best-practices

Best practices for HA automations, helpers, scripts, controls, and dashboards. TRIGGER THIS SKILL WHEN: - Creating or editing automations, scripts, scenes, or dashboards - Choosing between template sensors and built-in helpers - Restructuring triggers, conditions, or automation modes - Setting up Zigbee button/remote automations - Renaming entities or migrating device_id to entity_id - Configuring dashboard cards or selecting helpers - Looking up card types or domain docs - Writing or reviewing AppDaemon apps SYMPTOMS: - Agent uses Jinja2 templates where native options exist - Agent uses device_id instead of entity_id - Agent changes entity IDs without checking consumers - Wrong automation mode - Agent hard-codes values or uses raw sensor over helper - Agent edits .storage, writes YAML, or generates YAML snippets - Agent tells user to edit configuration.yaml for UI integrations

community