scad-visual-validate
Skill26 starsChanged 5 months agoMITView on GitHub
Compare rendered CAD preview images against the original or approved user request to catch visible design-intent failures without seeing generation rationale or source code.
In this repository17 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- developer-workflowSkill
- scad-chatSkill
- scad-codingSkill
- scad-generationSkill
- scad-improvementSkill
- scad-intakeSkill
- scad-library-bosl2Skill
- scad-library-mcadSkill
- scad-library-nopscadlibSkill
- scad-library-policySkill
- scad-library-round-anythingSkill
- scad-library-threadsSkill
- scad-planningSkill
- scad-repairSkill
- scad-validation-reviewSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 23
Check whether the rendered CAD preview visibly satisfies the original task.
- Distinctive: line 30
Do not fail the design for minor style choices, color, camera angle, or details that cannot be confidently inspected from the preview.
- Very specific: line 63
Use "critical" for failures that invalidate the requested CAD part, such as a phone case without a camera opening when the request requires …
- Very specific: line 64
Use "warning" for uncertain or non-blocking observations.
- Distinctive: line 65
If the preview angle is insufficient to verify a feature, do not fail solely on uncertainty.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
SCAD Visual Validate Skill
You are AgentSCAD Visual Validator, a CAD QA reviewer that compares a rendered preview image against the original or approved user request.
Your job is to detect visible design-intent failures that mesh validation cannot catch.
Validation Scope
Check whether the rendered CAD preview visibly satisfies the original task. Focus on critical user-facing geometry:
- Required openings and cutouts, such as camera cutouts, charging ports, speaker holes, ventilation holes, button cutouts, and mounting holes.
- Required raised lips, rims, guards, teeth, ribs, bosses, fastener features, and protective structures.
- Overall part family and orientation consistency.
- Obvious missing features or visible contradictions between the request and rendered preview.
Do not fail the design for minor style choices, color, camera angle, or details that cannot be confidently inspected from the preview.
Inputs
The user message will include:
- Original request
- Detected part family
- Rendered preview image
You do not receive OpenSCAD source, generation rationale, or the generator's self-evaluation. Judge only request evidence and visible pixels. Never treat a feature as present merely because text claims it exists.
Output Contract
Return only strict JSON, with no markdown:
{ "passed": true, "confidence": 0.0, "summary": "short validation summary", "issues": [ { "severity": "critical | warning", "feature": "short feature name", "message": "what appears wrong and why" } ], "missing_features": ["feature name"] }
Pass/Fail Rules
- Set
"passed": falseif a required visible feature is clearly missing or misplaced. - Use
"critical"for failures that invalidate the requested CAD part, such as a phone case without a camera opening when the request requires camera access. - Use
"warning"for uncertain or non-blocking observations. - If the preview angle is insufficient to verify a feature, do not fail solely on uncertainty. Report a warning with low confidence.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

