Manus Ships typst-pdf-maker Turning Plain Prompts Into Print-Ready PDFs

Getting a polished, print-ready PDF out of an AI agent has always required some assembly. You'd get the text, then wrestle with Word, LaTeX, or a design tool to make it look professional. Manus just collapsed that gap with a new built-in skill: /typst-pdf-maker. Describe what you want, and Manus hands back a fully typeset PDF document.

What's a Skill, and why does it matter here?

Manus Skills are modular, reusable workflows you trigger with a slash command. They are modular, file-system-based resources that encapsulate a specific capability or workflow -- think of them as a detailed "onboarding guide" you hand to your agent, enabling it to perform specialized tasks with precision and consistency. The key distinction from a plain chat prompt is execution: Manus actually executes your skills end-to-end in a secure VM and delivers completed results, meaning it can fully automate complex workflows, not just suggest how to do them.

Under the hood, Manus runs in an isolated VM with full Ubuntu filesystem access and shell execution, reading SKILL.md files and executing Python/Bash scripts to orchestrate workflows automatically. The /typst-pdf-maker skill plugs directly into that infrastructure, using the Typst compiler to render the final document.

Typst: the rendering engine doing the heavy lifting

If you haven't encountered it before, Typst is a modern typesetting system worth knowing. It is designed to be as powerful as LaTeX while being much easier to learn and use. The practical difference is dramatic: Typst produces pixel-perfect PDFs from structured markup in milliseconds, not seconds.

For automated generation specifically, Typst has real advantages over the alternatives. LaTeX offers beautiful typography but was designed for manually edited manuscripts, not automated generation -- ingesting data requires error-prone scripts or Lua scripting with a steep learning curve, and its distributions are large (over 1GB), making them ill-suited for serverless environments. Typst sidesteps all of that: it is a single lightweight (~40MB) and secure binary that is easy to deploy and run in containerized cloud environments.

Typst is designed for writing and formatting scientific texts and mathematical formulas, supporting simple formatting for common applications, customizable functions, an integrated scripting language, and mathematical typesetting. That scripting layer is what makes AI-driven generation practical -- Typst allows you to load data from any format, with parsers for XML, JSON, and CSV included.

From prompt to polished document

The workflow is straightforward. You type /typst-pdf-maker in the Manus chat, describe the document you need, and the agent generates Typst markup, compiles it inside its sandboxed VM, and returns a finished PDF. No intermediate steps, no manual formatting pass.

Manus says the skill is designed for:

• Research reports and technical papers
• Resumes and CVs
• Business proposals and briefs
• Full books and long-form documents

The use of Typst as the rendering layer means the output inherits professional typographic defaults -- proper heading hierarchies, consistent spacing, clean table layouts -- without you specifying any of it. Typst is a good choice for writing any long form text such as essays, articles, scientific papers, books, reports, and homework assignments.

Where it fits, and where it doesn't

This skill shines for structured, text-heavy documents where layout consistency matters more than pixel-level brand control. Think internal reports, academic-style writeups, technical documentation, or a resume that needs to look clean without you touching a template.

It's less suited for:

• Heavily branded marketing collateral (custom fonts, brand colors, precise visual identity)
• Documents requiring complex multi-column magazine-style layouts
• Anything needing live data pulled from external APIs at render time

It's also worth noting that Typst, while powerful, is still a relatively young ecosystem. For long-term support documents, such as documentation, it may be worth waiting to see it become more stable. For one-shot document generation, though, that caveat barely applies.

How to use it

Once a skill is added to your library, you can activate it at any time during a conversation by typing / in the chat input, which brings up a list of available skills -- select the one you want to use to instruct Manus to load its instructions and execute its workflow. For /typst-pdf-maker specifically, the more detail you give in your prompt (document type, audience, key sections, tone), the better the output.

A quick example prompt that would work well:

/typst-pdf-maker
Create a 4-page technical report on the tradeoffs between
transformer and state-space models for long-context tasks.
Include an executive summary, a comparison table,
and a recommendations section. Audience: ML engineers.

The skill is available now to all Manus users. Pricing runs from free (300 credits/day refresh) to Plus at $39/month, Pro at $199/month, and Team at $39/member/month. Free users also get 1,000 starter credits that never expire.

The bigger picture

The /typst-pdf-maker skill is a small but telling signal about where agentic AI is heading. The bottleneck in AI-assisted work has quietly shifted from content generation to output formatting -- getting the right words is now easy; getting a document that looks like a human professional made it is still hard. Manus is betting that the answer is to wire a proper typesetting compiler directly into the agent loop, rather than dumping markdown into a generic PDF renderer. Given Typst's adoption trajectory -- Zerodha adopted Typst for generating transaction records, and scientists working on SciPy conference proceedings adopted it for PDF generation -- the underlying technology is already battle-tested in production pipelines. Bringing it to a prompt-driven interface is the logical next step.