SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsartcraft

artcraft

Design & creation

View repository

Updated 10-29 04:35

About this skill

When to use this skill

Use this skill for source-grounded ArtCraft desktop operation: fit checks, offline preflight, release-versus-source decisions, 2D/3D scene blocking, provider and credit disclosure, approved AI generation, export/download handoff, and troubleshooting. It covers the repository at the pinned commit, not an unverified moving checkout.

Instructions

  1. Run the offline gate first. From the repository root run python3 .agent-skills/artcraft/scripts/preflight.py --json; from an installed skill root run python3 "$SKILLS_ROOT/artcraft/scripts/preflight.py" --json. Add --repo PATH --expect-commit 40e04793b504147de97ba1f0659d3975167524e5 only for a supplied checkout. The helper is read-only and never starts ArtCraft, Tauri, Node, providers, servers, or network work.
  2. Fit-check the path. Read references/upstream.md, then distinguish a signed desktop release for Windows/macOS from a source build that can include Linux. Treat a missing checksum or signature as an unresolved verification gap, not as proof of authenticity.
  3. Install a release only after approval. Confirm target OS, release URL, license/NOTICE obligations, and whether the upstream release provides a checksum or signature. Never run an installer, download an artifact, or bypass OS trust prompts without explicit approval.
  4. Build from source only after approval. Use the pinned source and the documented Rust, Node.js 20+, npm, and Tauri CLI toolchain. Do not run npm ci, cargo tauri, cargo check, or any build command until approved; source builds use the normal ArtCraft data and Cargo target locations.
  5. Block scenes deliberately. In the app, choose a 2D canvas or 3D stage, place image layers, props, meshes, or kitbashed assets, establish camera and depth, then preserve the blocking state before generation. Route Blender-owned scene inspection or manipulation to blender-mcp.
  6. Gate every AI generation. Before image, video, audio, mesh, or world generation, disclose the selected provider/model and the estimated credit or USD cost when the UI reports it. Require separate approval for login, API keys or cookies, remote calls, uploads, and paid credits; ArtCraft's hosted services are not local inference.
  7. Export and hand off explicitly. Use the app's download/export controls and record the destination, media type, model, and job result. Treat downloaded media and remote CDN links as user data; do not publish, share, or upload them without approval.
  8. Troubleshoot without widening scope. Capture app version, OS, source pin, exact failing mode, and sanitized logs. Check the pinned platform and data-directory notes before resetting state; clearing credentials, tasks, settings, or downloads is destructive and requires approval.
  9. Report evidence. Separate source facts, observed local checks, estimates, and runtime results. State when no generation or build was run, and do not claim a model, release, checksum, or provider is available beyond the pinned evidence.

Examples

Run the ArtCraft offline preflight against this checkout, compare release versus source-build options for macOS, then outline a 3D blocking workflow. Do not install, build, sign in, call a provider, or spend credits.

Best practices

  • Keep scene blocking and generation as separate checkpoints so a failed remote job cannot erase composition work.
  • Display provider, model, credit estimate, and approval state before any enqueue or upload action.
  • Preserve the pinned commit, app version, selected model, and export path in the handoff record.
  • Treat credentials, local task databases, settings, downloads, and telemetry as sensitive operational data.
  • Prefer open-generative-ai for a hosted multi-provider studio, openstory for script-to-video storyboards, blender-mcp for Blender control, and video-production for code-first assembly.

References