ReportKit

ReportKit exports prepared Markdown or report.json content into polished Typst PDF reports.

It is designed for LLMs and agents at the final delivery step: the agent prepares the content, then calls report-kit to produce a stable PDF.

Install

# macOS Apple Silicon
npm install -g @dztabel/reportkit @dztabel/reportkit-darwin-arm64

# Linux x64
npm install -g @dztabel/reportkit @dztabel/reportkit-linux-x64

# Windows x64
npm install -g @dztabel/reportkit @dztabel/reportkit-win32-x64

The canonical command is report-kit. reportkit is also provided as a compatibility alias.

Check the command:

report-kit --version

Quick Start

Create content.md:

---
title: Project Review Report
subtitle: Agent-prepared content
author: ReportKit
date: 2026-05-26
---

# Executive Summary

This report was prepared by an upstream agent. ReportKit is responsible for turning the prepared content into a polished PDF.

表：Delivery Checklist
| Item | Status | Notes |
| --- | --- | --- |
| Content prepared | Done | The agent has already written the report. |
| PDF export | Ready | ReportKit handles layout and rendering. |

Export the report:

report-kit build content.md --out ./report

Output files:

report/report.pdf
report/report.json
report/report.typ
report/<template-id>.typ
report/render-result.json
report/build-result.json

Inputs

ReportKit accepts:

• Markdown prepared by an LLM or agent.
• A structured report.json file.

For direct report.json input, formula fields are treated as Typst math only and must not contain Typst code escapes such as #read(...). Figure paths are resolved through an asset allowlist rooted at the input file directory, output directory, or built-in example assets; arbitrary absolute local file paths are rejected.

It is useful when the user asks for a formal PDF report, consulting-style deliverable, project review, research summary, or Chinese business report.

What ReportKit Does

• Converts prepared content into a report structure.
• Applies a stable Typst report style.
• Generates PDF output.
• Returns machine-readable build results (errors + warnings) for agents; dangling cross-references fail the build with the actual numbering listed.
• Auto-numbers tables/figures/equations per chapter and renders prose references (见表 x.x / 如图 x.x / 式 x.x) as clickable cross-references.
• Renders a final sources chapter as numbered [1] [2] reference entries with clickable superscript [n] citations in prose.
• Resolves table column widths and font size from measured content (descriptive columns widen, short columns stay compact).
• Keeps table captions, figure captions, formulas, code blocks, lists, explicitly requested callouts, and long tables in a controlled layout.

What ReportKit Does Not Do

• It does not search the web.
• It does not judge facts.
• It does not replace the upstream LLM or agent's writing process.
• It does not require users to write Typst by hand.

Agent Usage

An agent can call:

report-kit build prepared-report.md --out ./report

Then read report/build-result.json or stdout to locate the generated PDF.

Successful builds return JSON with ok: true, artifact paths, component counts, and warnings.

Platform Support

Current public beta:

• macOS Apple Silicon: supported through @dztabel/reportkit-darwin-arm64.
• Linux x64: supported through @dztabel/reportkit-linux-x64.
• Windows x64: supported through @dztabel/reportkit-win32-x64.

More platform packages can be added without changing the main @dztabel/reportkit command.