Java LLM Extractor

A pragmatic utility designed to crawl through a JSF/Java legacy application and consolidate its menu architecture into a structured, LLM-ready Markdown document.

This tool was built to solve the "context window" problem for AI agents (Copilot, Cursor, Claude, etc.) working on large-scale legacy systems — especially those built with JSF + PrimeFaces + Java EE. Instead of feeding thousands of raw source files into an LLM, you feed it the relevant slice of this document.

Core Principles

• Performance over contraptions: Scans thousands of Java files in seconds, resolving labels, URLs and business methods in a single pass.
• Operational efficiency: Single-binary execution. No Node.js runtime required on the target machine.
• KISS: No complex UI. Just a CLI that does one thing: turns a JSF sidebar + Java source tree into a clean, semantic Markdown map.

What it extracts

For each item in the application menu (sidebar.xhtml), the tool produces:

• Menu hierarchy — module > submodule > function (up to 3 levels)
• Screen URL — the JSF page path
• Permission key — the access control identifier (trinityUtils.userHasPermission)
• Java classes — the backing beans and controllers mapped to that screen
• Business operations — public non-getter/setter methods found in each class
• Permission index — a full table of all permissions cross-referenced to their function and module

Usage

Development (requires Node.js 18+)

npm install
npx tsx index.ts "<project-root>" "<output-file>" [--utf8]

Arguments:

Argument	Description	Default
<project-root>	Root path of the Java project	current directory
<output-file>	Output Markdown file name	menu-architecture.md
--utf8	Write output in UTF-8 instead of ISO-8859-1	Latin-1 (default)

Example:

# Latin-1 output (standard)
npx tsx index.ts "C:\Projetos\Company\Company-1.0-master" "menu-architecture.md"

# UTF-8 output (for LLM APIs or web tools)
npx tsx index.ts "C:\Projetos\Company\Company-1.0-master" "menu-architecture.md" --utf8

Binary (no Node.js required)

Download the binary for your OS from the GitHub Actions tab (under Artifacts) or build it locally:

npm run build

Then run:

# Windows
.\bin\java-llm-extractor-win.exe "C:\Projetos\Company\Company-1.0-master" menu-architecture.md

# Linux
./bin/java-llm-extractor-linux "C:/Projetos/Company/Company-1.0-master" menu-architecture.md

Expected project structure

The tool expects a standard Maven multi-module layout:

<project-root>/
  web/src/main/
    webapp/layout/sidebar.xhtml     ← menu tree
    properties/messages.properties  ← PT-BR labels
    java/...                        ← backing beans
  core/src/main/java/...            ← core services
  common/src/main/java/...          ← shared utilities

Output sample

##### Pesagem de Entrada

- **URL:** `/com/arcadian/product/web/pesagemEntradaModalRodoviario/pesar-modal-rodoviario-na-entrada.jsf`
- **Permission:** `pesar-modal-rodoviario-na-entrada`
- **Java Classes:**
  - `web\src\main\java\...\PesagemEntradaModalRodoviarioController.java`
    - Operations: iniciar, confirmar, cancelar, pesarVeiculo, gerarTicket

Encoding

Source files are read in ISO-8859-1 (Latin-1), which is the encoding standard. The output file is written in the same encoding by default. Use --utf8 to produce a UTF-8 output when sending the document to external LLM APIs or web-based tools.

Artifacts:

• Binaries

Author

Andre Mesquita