Describe logo

Describe

Organization
partcad
describe

Write a narratable, accessibility-oriented text description of an existing PartCAD part, assembly, or sketch from its measured bounding box and volume, from renders taken at several viewing angles, and - where it is available - from a dimensioned technical drawing. Use for /pc:describe or when the user asks to describe, summarize, or caption a PartCAD object. Reproduces the retired built-in AI shape-summary.

Overview

Publisherpartcad
Repositorypartcad
Skill namedescribe
Stars
504
Forks
47
Bundled files
Instructions only
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by partcad on GitHub. Read the source before you install it.

Installation

Install the Describe AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/partcad/partcad/tree/devel/ai-agents/common/skills/describe
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/partcad/partcad.git /tmp/partcad
mkdir -p .claude/skills
cp -r /tmp/partcad/ai-agents/common/skills/describe .claude/skills/describe
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Describe in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Describe on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Describe is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

pc:describe

Produce a text description of an existing PartCAD shape by rendering it and looking at the result — the replacement for PartCAD's retired built-in AI summary. The text after the command ($ARGUMENTS) names the object to describe (optionally with its kind).

One picture is not enough to describe a shape. A single projection hides everything behind it, and a description written from one is a description of a silhouette. Render three views (§3), and where a dimensioned drawing can be had, read the numbers off that rather than estimating them (§4).

Nor is any picture enough to give a size. A projection is rendered to fit its frame, so nothing in it says whether the part is 20 mm across or 200 — which is why this starts by asking PartCAD what it measured (§2) and never estimates a dimension a number was available for.

1. Resolve the object

$ARGUMENTS is the object name (a part by default; an assembly or sketch if the user says so). Read its desc:/requirements: from partcad.yaml for context. Make sure PartCAD is available as /pc:init describes.

Pass --no-ansi on every run so the output is plain text; it is a global flag, goes before the subcommand, and routes the logs to stderr.

This skill describes an object a package declares, because §6 stores the result back into that package. To describe a bare CAD file that is in no package, render it with /pc:render (pc adhoc render) and write the description for the user instead of storing it.

2. Measure it

pc info reports what PartCAD measured when it built the object, and those are the numbers this description is anchored on:

sh
pc --no-ansi info <name> 2>&1   # add -a / -s for an assembly or a sketch

Read all of it. The measurements are the first thing to look for, but the sections below are what the design says about itself, and a filter narrow enough to catch BoundingBox would cut them off.

  • BoundingBox — min, max and size in millimetres. size is the overall envelope of the design: state it, and give every other dimension in proportion to it rather than in pixels.
  • Volume — how much material is in it, in cubic millimetres, with Solids saying how many solids that is the volume of. More than one solid in a part is worth saying out loud: it is usually either deliberate (a part that is two pieces) or a mistake nobody has noticed. A negative volume means the faces are oriented inward — report that as a defect rather than dropping the sign.
  • A sketch, a shell or a wire has no volume, and pc info says so by leaving Volume out. It is flat, so its bounding box is its size.

The rest of what pc info prints is worth a look too, and for an object read from a file it is where the design's own words are: a STEP file's File, Products, Layers and Properties, a DXF sketch's Drawing, Layers and Annotations — including the angle, radius and direction of a sheet metal bend. Anything stated there is the design's own and beats a reading off a picture.

The first run builds and measures the object, so it is slow; every run after it is a cache hit. If it fails — no sandbox, a part that does not build — say so in the report and describe the shape from the views alone, without inventing a size.

3. Render three views

front, top and iso: two orthographic views to read proportion and features from, and one pictorial view that shows how they go together. A rendered file is named after the object, so give each view a directory of its own or they overwrite each other:

sh
for view in front top iso; do
  mkdir -p /tmp/pc-render/$view
  pc --no-ansi render -t png --view $view -O /tmp/pc-render/$view <name>   # add -a for an assembly
done

View all three — /tmp/pc-render/<view>/<name>.png.

Add right (or back, left, bottom) when the three leave a face unaccounted for: a part that is not symmetric about an axis has a side you have not seen, and guessing at it is exactly what this skill must not do.

A sketch is flat, so it has only one view worth rendering, and SVG suits it better than PNG:

sh
mkdir -p /tmp/pc-render
pc --no-ansi render -s -t svg -O /tmp/pc-render <name>

4. Render a dimensioned drawing (parts, when it is available)

The projections above show shape but no numbers. draftwright, a render implementation published in the public PartCAD index, draws a fully dimensioned technical drawing instead — orthographic views with dimension lines, hole callouts, radii, an ISO view, and a title block carrying the material and scale. Reading a description off that beats estimating from pixels:

sh
mkdir -p /tmp/pc-render/drawing
pc --no-ansi render -t pdf -e //pub/feature/render/draftwright \
    -O /tmp/pc-render/drawing <name>

-e names the package to read the output configuration from, so this works on any object without that object's package knowing about draftwright. Then read /tmp/pc-render/drawing/<name>.pdf — ask for pdf, whose pages can be looked at directly. Its svg draws every character as paths, so no text can be read out of that markup.

This is a best-effort extra, not a requirement. It needs the public index among the package's dependencies: (which pc init writes) and a network connection, and the first run is slow while PartCAD installs draftwright into a sandbox. It is also a part drawing — do not expect it of an assembly or a sketch.

draftwright reports what it could not do. When a part has features it cannot dimension at the chosen scale it says so and draws the rest, and the command may exit non-zero having still written a usable file — so check whether the file appeared before giving up on it. If it did not, or anything above failed, describe the object from the three views and say in your report that no dimensioned drawing was available.

5. Write the description

From the measurements, the images and the configured desc/requirements, write a description for a reader who has a mechanical-engineering / CAD background but cannot see it:

  • State every measurement §2 produced, and say so when it produced none. The overall envelope in millimetres, and for a part how much material is in it: those are measurements rather than impressions, and they are what tells a reader whether this is a bracket or a building. Where pc info gave no such number — a sketch, which has no volume; a part that would not build; a machine with no sandbox — write that the measurement was unavailable. Saying so costs a clause; the alternative is a figure you made up, which is the one thing this skill exists to prevent.
  • Describe both the overall shape and the dimensions; make no assumptions and add as much concrete detail as the renders and config support.
  • Prefer a measured or stated dimension over one you judged from a projection: pc info first, then the dimensioned drawing, then the picture. Where you are reading a proportion off a picture rather than a stated number, say it is approximate rather than inventing a figure.
  • Work through the views: overall form and bounding size first, then the features each view reveals, then how they relate.
  • Refer to it as "this design", not "the image" — and never as "the three images", which is an artifact of how you looked at it.
  • Produce prose that is ready to be narrated as-is (no markdown, no lists).

6. Persist it (so pc ide view shows it)

pc ide view reads a shape's summary: from its configuration. Write your description there so the tool and other agents can reuse it:

yaml
parts:            # or sketches: / assemblies:
  <name>:
    ...
    summary: >-
      <your description>

Report the description to the user, say which views (and whether a dimensioned drawing) it was written from and which numbers were measured rather than judged, and confirm where you stored it.

Frequently asked questions

What does the Describe AI skill do?

Write a narratable, accessibility-oriented text description of an existing PartCAD part, assembly, or sketch from its measured bounding box and volume, from renders taken at several viewing angles, and - where it is available - from a dimensioned technical drawing. Use for /pc:describe or when the user asks to describe, summarize, or caption a PartCAD object. Reproduces the retired built-in AI shape-summary.

Why use Describe on TypingMind?

Because you install it once and use it with any model. Describe is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Describe in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/partcad/partcad/tree/devel/ai-agents/common/skills/describe. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Describe?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Describe?

As many as you like. As long as a model supports skills, you can use Describe with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Describe AI skill free?

Yes. It is published on GitHub by partcad under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇