tex-render

name	tex-render
description	Renders LaTeX math to PNG, JPEG, WebP, or AVIF images using MathJax (TeX→SVG) and @svg-fns/svg2img. Invoke whenever the agent needs to output LaTeX as a viewable image (equations, formulas, notation).
metadata	{"openclaw":{"emoji":"📐"}}

TeX Render 📐

Renders LaTeX math to PNG, JPEG, WebP, or AVIF (and SVG). Use when you need a viewable image from LaTeX instead of raw code.

User notice: When this skill is active, the agent will automatically render any LaTeX in its replies as images and send them in order—without asking for permission. If you prefer to be prompted or to receive raw LaTeX instead, do not enable this skill (or remove it from your workspace).

Location

The render script lives in the same skill folder as this SKILL.md:

<skill_folder>/
├── SKILL.md
├── package.json
└── scripts/
    ├─── render.js
    └─── validate.js

Use the directory containing this SKILL.md as the skill path. The script is at scripts/render.js relative to that folder. Invoke: node <skill_folder>/scripts/render.js. The agent loading this skill has the skill path—use it to run the script.

Install

One-time setup. After clawhub install tex-render, run npm install in the skill folder:

cd <skill_folder>
npm install

• Source: All dependencies are from the public npm registry only (no ad-hoc downloads, shorteners, or remote archives).
• Native module: The sharp package may compile during install; ensure Node.js 14+ and a working build toolchain if needed (see sharp install for platform notes).

When to Use

• User or task asks to "render this equation as image" or "show formula as picture"
• Your reply would contain LaTeX — render formulas as images first, then respond with plain text

Workflow: Response Contains LaTeX

Output your reply interleaved: whenever you would output LaTeX, do this instead:

1. Send the preceding plain text — call message with the text written so far (no LaTeX).
2. Render the LaTeX expression with this skill (default PNG; no --output dataurl). Parse the JSON for the PNG path.
3. Send the image — call message with action: "send", path set to the rendered PNG, and message as short caption.
4. Continue outputting the rest of the message. Repeat the cycle (text → render → send image) for each LaTeX block.

Do not output raw LaTeX. Do not ask the user for permission to render — render and send images immediately when LaTeX would appear in your reply (users are informed of this automatic behavior in the user notice above). Do not accumulate everything and send at the end — send text and images in order as you go.

Critical: Output in order: plain text → send → LaTeX → render → send image → plain text → send → LaTeX → render → send image → … The user must receive text and images in the natural reading order.

Example: Explaining Lagrangian:
→ Send "The Lagrangian is defined as " → Render L = T - V → Send image with caption "L = T - V" → Send ". The Euler-Lagrange equation is " → Render d/dt(∂L/∂q̇) - ∂L/∂q = 0 → Send image → Send " — this yields the equations of motion."

Usage

Run: node <skill_folder>/scripts/render.js (or node scripts/render.js when already in the skill folder).

Escaping: When invoking via shell:

• Single quotes for LaTeX without apostrophe: node scripts/render.js '\frac{d}{dt}'
• Stdin when LaTeX contains apostrophe (e.g. y' for derivative): printf '%s' "y' = f(t, y), \quad y(t_0)=y_0" | node scripts/render.js — use double quotes so ' is literal; backslashes like \quad stay intact.

node <skill_folder>/scripts/render.js --help

Default output is PNG to ~/.openclaw/media/tex-render/. The script prints one JSON line with file paths: {"svg":"...","png":"..."} or {"svg":"...","jpeg":"..."}, etc. Use --output dataurl only when the conversation system explicitly supports Data URL images (otherwise it may show raw base64 text).

Examples (validated by npm test)

Use <skill_folder> = the directory containing this SKILL.md.

Basic (PNG default):

node <skill_folder>/scripts/render.js 'E = mc^2'
node <skill_folder>/scripts/render.js '$$\frac{F}{m}=a$$'

LaTeX with apostrophe (e.g. y'): use stdin to avoid shell quoting issues:

printf '%s' "y' = f(t, y), \quad y(t_0)=y_0" | node <skill_folder>/scripts/render.js

JPEG / WebP / AVIF:

node <skill_folder>/scripts/render.js --format jpeg --quality 80 '\frac{F}{m}=a' ./out/formula
node <skill_folder>/scripts/render.js --format webp 'x^2 + y^2 = z^2'
node <skill_folder>/scripts/render.js --format avif 'E = mc^2'

Data URL (no file):

node <skill_folder>/scripts/render.js --output dataurl 'E = mc^2'

Scale by width:

node <skill_folder>/scripts/render.js --width 800 '\int_0^\infty e^{-x^2} dx'

Inline math (smaller rendering):

node <skill_folder>/scripts/render.js --inline 'a^2 + b^2 = c^2'

Height and zoom: Use --height N or --zoom N as documented in --help.

Automatic Triggering (TOOLS.md)

To make the agent use tex-render without the user explicitly asking, add to your workspace TOOLS.md:

## LaTeX in Responses (tex-render)

**Whenever your reply would contain LaTeX** (equations, formulas, scientific notation), **use the tex-render skill** automatically. Examples: physics, math, chemistry questions with formulas.

**Workflow:** send plain text → render LaTeX → send image via message tool → continue text. Use single quotes when invoking (e.g. `'\frac{a}{b}'`).

### LaTeX / Equations (tex-render)

- When answering scientific or math questions, if your reply would contain LaTeX, use tex-render and send images — do this automatically.

Test: Ask "Explain the Lagrangian formula" without mentioning tex-render. The agent should render and send images.

Repository

This package is maintained at https://github.com/TheBigoranger/tex-render. You can open issues there for bug reports and feature requests.