Skip to content
Back to skills

Visualization Geometry

ASecurity

"Use for IKPy transformation-matrix checks, geometry helpers,

  • 247 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 8, 2026
toolsapibackend

Works with

  • cli
  • api

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 8, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill visualization-geometry --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Visualization Geometry?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Visualization Geometry
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-visualization-geometry/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-visualization-geometry)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: visualization-geometry
description: "Use for IKPy transformation-matrix checks, geometry helpers,
  headless 3D chain plots, target and frame overlays, and URDF tree
  visualization; do not use it to build chains, solve FK/IK, or control
  hardware."
disable-model-invocation: true
metadata:
  disco-role: operating
license: Apache 2.0
---

# Visualization and geometry

Use this sub-skill when a Researcher needs to inspect IKPy transforms or create a
visual diagnostic of a chain without connecting to a robot. Start with the
matrix and shape rules in [the API reference](references/api-reference.md),
then select a procedure from [the workflows](references/workflows.md). Keep
long-running or repeatable plots headless and save them to an explicit local
output path.

## Route by responsibility

- Route chain construction, link selection, forward kinematics, inverse
  kinematics, and joint-vector semantics to `chain-kinematics`.
- Route URDF/MJCF model loading and parser behavior to `robot-model-import`.
  This sub-skill only explains the `get_urdf_tree` visualization utility after
  a model path and root link are already known.
- Use the bundled `scripts/smoke_plot.py` for a dependency and save-path smoke
  check. It uses an inline toy chain and never reads a robot file or talks to
  hardware.
- Treat rendered images and DOT/PDF output as diagnostic evidence, not as a
  motion command, simulator state, or proof that a physical robot is safe.

## Operating sequence

1. Establish whether the requested result is a numeric transform inspection, a
   Matplotlib image, or a Graphviz tree; they have different optional
   dependencies and failure modes.
2. Check matrix dimensions and the RPY multiplication order before rendering.
   Inspect the numeric frame first, then use a plot to make frame placement or
   target error easier to see.
3. In headless environments, select the Matplotlib `Agg` backend before
   importing `ikpy.utils.plot`; call `Chain.plot(..., show=False)` and save the
   retained figure explicitly.
4. For a URDF tree, provide an existing URDF path and an exact root link name;
   inspect the returned DOT/tree objects before asking Graphviz to render.
5. Record output paths, dependency availability, numerical checks, and any
   clipping or convention assumptions in the downstream experiment log.

See:

- [references/api-reference.md](references/api-reference.md) for public
  functions, return shapes, conventions, and optional dependencies.
- [references/workflows.md](references/workflows.md) for transform checks,
  headless plots, target/intermediate-frame overlays, and DOT rendering.
- [references/troubleshooting.md](references/troubleshooting.md) for import,
  backend, shape, plotting, and safety failures.

Files in this skill

  • SKILL.md2.7 KB
  • references/api-reference.md7.7 KB
  • references/troubleshooting.md6.7 KB
  • references/workflows.md7 KB
  • scripts/smoke_plot.py3.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.