Skip to content

armgit / kiln

Public Imported from GitHub
0

Build and revise procedural 3D assets with your coding agent. Local MCP server, CLI and TypeScript engine with rendered review, editable source and GLB export.

2.7 KB
Raw Blame

Headless asset generation

The repository's dispatcher runs a coding-agent harness in a temporary authoring directory, then builds the resulting program independently. It keeps the brief, transcript, source, GLB, views, and result metadata for review.

node scripts/dispatch-asset.mjs --harness opencode --model PROVIDER/MODEL --name water-tower "A steel water tower on a riveted lattice frame"

Use a model available through your configured provider. For an OpenCode model, confirm image input support with node scripts/check-vision.mjs PROVIDER/MODEL, which reads the installed OpenCode's model catalogue; for other harnesses, use the provider's model documentation. Then check that the harness actually forwards render images. Provider metadata alone does not verify the whole path.

Run bun run smoke:harness -- --harness opencode before a batch. A quota or authentication failure is not a measure of asset quality. Inspect the retained transcript and build result when a run fails.

Per-harness install, upgrade, headless flags and MCP config locations are in headless harnesses. For new integrations, start with the workspace setup guide. The older dispatch adapters in scripts/harness.mjs have their own harness-specific setup and may inherit user configuration; do not assume every adapter provides the same isolation as a fresh workspace.

Comparing runs

Keep the brief, supplied files, renderer, and time allowance consistent. Record the exact model and harness version, whether images reached the model, and any inherited context. For refinement, supply only the source being edited; for new assets, omit the example library and engine source.

Review silhouette, construction, attachments, ground contact, and requested detail. Judge materials only from material-faithful images. A valid GLB, a high triangle count, or the model's own success message does not establish that the brief was met.

An agent that delegates the authoring to a nested agent must start that agent as a separate harness process in the workspace directory. For OpenCode 2, change into that directory and run opencode run --standalone "<brief>". A subagent of the outer session inherits the outer session's tools and never loads the workspace's MCP configuration, so it is left with the CLI alone. The tier-2 driver's receipt records this as toolUsage.workspaceMcp: exercised, not-exercised or unknown.

Use separate workspaces for independent runs and choose concurrency within the available budget. Keep interrupted artifacts, but mark them as partial when they do not satisfy the request. See the batch-dispatch skill for collection and provenance guidance.