pinepaper
GitHub用于构建和检查Pinepaper交互式图表、动画及物理模型,支持教学场景中的动态可视化需求。
Trigger Scenarios
Install
npx skills add madhvantyagi/Gnos --skill pinepaper -g -y
SKILL.md
Frontmatter
{
"name": "pinepaper",
"description": "Build and check Pinepaper MCP diagrams, charts, motion, and interactive models when a lesson needs linked visual states or a changing quantity."
}
Pinepaper
Use the pinepaper MCP server for a lesson view whose parts must stay linked or
change together. Start with the selected subject guide and the block brief.
Name the question the learner should answer from the scene. Keep the same
objects, symbols, units, and example as the surrounding lesson.
Choose the visual job
- For a fixed relationship or boundary, read diagrams. Use Excalidraw when a quick static sketch is enough.
- For a plotted quantity, curve, or linked numerical view, read charts and quantities. Keep exact values in a table or reproducible calculation.
- For ordered states, physical motion, waves, or a model such as heat transfer, read motion and models.
- For learner controls or any finished file, read interaction and export. Choose the export before building features that the format cannot preserve.
One lesson may combine a diagram, graph, and motion around the same state. Each view must reveal a different part of the explanation. Do not make the learner infer which curve, arrow, or moving object corresponds to a number in the text.
Build through the MCP server
- Write a short scene brief: claim, objects, source of quantities, units, what changes, and the expected output file. Decide what the learner will predict or compare.
- Read the current tool schema. Use
pinepaper_tool_guidefor a tool family orpinepaper_query_capabilitiesfor a specific missing capability. The server schema wins over examples in these files. - Call
pinepaper_agent_start_jobwith the canvas preset and a description of the teaching claim. Plan positions and labels at the size the learner will see in the course viewer. - Use one
pinepaper_agent_batch_executefor the base scene when its operations cover the job. Reuse its$0,$1, and later item references. Call a specialized tool, such aspinepaper_create_chart,pinepaper_equation_path, orpinepaper_physics, when the scene needs that capability. Do not approximate a chart or physical rule with decorative movement. Check the returned result of every operation. - Run
pinepaper_validate_scene. Inspect the initial state and a meaningful changed state with a screenshot or playback. Check labels, scales, arrow meanings, clipping, and whether the graph and motion agree. Repair failed operations or unwired relations before finishing. - Call
pinepaper_agent_end_job, inspect its final image, then export throughpinepaper_agent_exportor the specialized export tool required by the deliverable. Open the exported file. A correct editor scene is not proof that its export works.
Use named tools for ordinary production. Use pinepaper_execute_custom_code
only for a bounded scene behavior the tool surface cannot express, and check
the generated values against the lesson's model. A visually plausible model
is still a model. Label measured data, sourced data, assumptions, and simulated
outputs distinctly.
Hand off a checked artifact
Return the file, MIME type, title, purpose, source or model assumptions, and one learner question to the lesson coordinator. The coordinator alone updates the lesson and registers the artifact. Check the export in the course viewer before calling it ready. If the server is unavailable, use its diagnostic report when available, make one error-based correction, and report the remaining limitation. Never describe an unmade scene as finished.
Version History
- c0bef66 Current 2026-09-27 10:19


