Software

Open-source Agent Skills that keep scientific work auditable, from optical tables to measurement software and literature reviews.

Featured project · Agent Skill · v2.0.0

OpticalModeler

Light. Structure. Evidence.

Turn measurement ideas, diagrams and photographed setups into editable Blender optical systems, tied to real apertures, mounts and support paths, with the saved scene, checks and final views kept on one evidence trail.

137 stars
Python
Blender
MIT license

View on GitHub

npx skills add k-telux/OpticalModeler

Choose the work, preserve its boundaries.

Five kinds of request, each starting from what may and may not change.

Design a new measurement

An empty scene, a source-backed topology and permitted component assets.

Reconstruct a diagram or photographs

Authoritative inputs and an explicitly retained baseline.

Correct an accepted model

Frozen originals, protected families and declared movable degrees of freedom.

Audit an existing scene

Read-only evidence, actual saved geometry and precise findings.

Present the accepted result

A separate camera and light copy, with geometry and image provenance preserved.

The optical path.

Directed branches, actual working faces, apertures and detector endpoints. Free-space light and guided fiber keep distinct roles.

The physical structure.

Real mounting interfaces, table holes, fasteners and continuous supports, checked in the saved scene against their neighbors.

The evidence.

One scene lineage, fresh reopen measurements, actual image producers and a consistent manifest behind every claim.

A public 2D optical schematic rebuilt by OpticalModeler as a complete Blender optical table with colored beam paths — **From a public schematic** to a complete optical table.

Annotated top view of the reconstructed optical table — **Annotated top view** of the same table.

Cutaway render from an optical coherence tomography forward test — **OCT forward test,** cutaway view.

Know what the result establishes.

PASS: The declared applicable gates have current evidence.
PARTIAL / SCOPED: A stated subset is checked; remaining blockers stay visible.
UNVERIFIED: Necessary evidence is missing or inconclusive.
BLOCKED: A known requirement fails.

Independent community workflow, not affiliated with or endorsed by Thorlabs. A ready-for-review handoff does not certify installed hardware, alignment, optical performance or laser safety.

More Agent Skills.

The same evidence-first approach, applied to lab software and scientific writing.

1 · Python · MIT

Labview2Python

Migrate LabVIEW measurement software to Python with simulator, UI, hardware, data-parity and packaged-EXE proof.

LabVIEW
PySide6
Lab automation
Hardware control

1 · Python · MIT

SciReviewWriter

Mechanism-first, citation-traceable scientific reviews with LaTeX, rendered-page QA and multilingual rules.

Literature review
LaTeX
Citations
Codex plugin

From the lab bench.

All repositories

HTML · Python

Software

OpticalModeler