More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    GraphCompose AI Flow: Reference to Java PDF Template

    1

    GraphCompose AI Flow is a harness for Claude Code and Codex that turns a document reference into a maintainable Java template and proves it matches.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    GraphCompose AI Flow: Reference to Java PDF Template

    GraphCompose AI Flow: Reference to Java PDF Template

    Example session with this skill installed

    Create a GraphCompose invoice template from invoice.png in my Java project at ./billing-service.

    • Read your context and instructions
    • Compiled the graphcompose ai flow

    Template ready for approval: invoice (revision-004)

    Example of the end of a create run. The skill ran preflight, imported the reference and went through four render-and-diff passes.

    PassWhat changedResult
    revision-001First authoring: header, bill-to, line items, totals, footerrendered, 6 regions over threshold
    revision-002Line-item table: column weights and row padding3 regions over threshold
    revision-003Header: logo resolved from assets, title typeface matched1 region over threshold
    revision-004Totals block aligned to the table's right edgeall regions within threshold

    Files in your project (graphcompose-flow/projects/invoice/revisions/revision-004/)

    • generated-template.java: the template, built from sections, rows and anchors, no raw coordinates
    • invoice-data.json: the content, behind a typed spec, so a new invoice is a new JSON file
    • output.pdf / output.png: the render
    • visual-review.md: per-region diff results and what each pass fixed

    Status: READY_FOR_APPROVAL

    Next steps

    1. Open output.pdf and compare it with your reference.
    2. Say "approve" to publish the template bundle under graphcompose-flow/templates/, or ask for a change ("make the header navy", "use Lato"), which opens revision-005.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Recreating a document with an AI agent usually ends in one of two places: a screenshot-shaped pile of absolute coordinates nobody can maintain, or "looks about right" with no way to prove it.

    GraphCompose AI Flow is a harness for Claude Code and Codex that turns a document reference into a maintainable Java template and proves it matches. Give it a reference (a screenshot, PDF or design image of a CV, invoice, proposal, cover letter or report) and it produces a maintainable Java template built from semantic GraphCompose primitives: sections, rows, weights and anchors, with the content in a JSON data file behind a typed spec.

    How it works

    1. Analyse the reference: page size, regions, typefaces, colours, assets.
    2. Author GraphCompose code against a pinned, closed API allow-list, so the agent cannot invent methods that do not exist.
    3. Render and diff: the template is compiled, rendered to PDF/PNG and measured against the reference, region by region.
    4. Fix one cause per pass and repeat, within a bounded loop, until the result is ready for your approval.
    5. Approve: the revision is published as a reusable template bundle in your project.

    Every change is a new revision; an approved one is never overwritten. Parity is a gate result from the diff tool, not the agent's opinion.

    What you get

    • Four workflow skills: create, revise, review and approve a template
    • Deterministic tools: renderer, pixel diff, region analysis, structural checks
    • Knowledge packs for GraphCompose 1.6 to 2.4 (API surfaces generated from the compiled library)
    • A first-run bootstrap that checks Java 21, Maven and ImageMagick, installs them with your consent, fetches GraphCompose from Maven Central and registers the skills with Claude Code or Codex

    Good to know

    • Output is Java: you need a Java project (Maven or Gradle) that uses GraphCompose, or a new one.
    • Best results today are in Claude Code; Codex is fully supported.
    • Open source under MIT: github.com/DemchaAV/graphcompose-ai-flow

    Requires Java 21+, Maven, and ImageMagick. Operates only on GraphCompose API surface. Only handles document templates (CVs, invoices, etc.). Needs network for bootstrap/Maven.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 21 days ago

    Requires

    Node.js 20+
    Java 21+ (JDK)
    Apache Maven
    ImageMagick 7
    macOS
    Node.js
    Git

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    Shell: runs node, mvn and magick to render and compare documents. Read/write files: writes templates and revisions into graphcompose-flow/ inside your Java project. Network: npm and Maven Central on first run, to install runtime dependencies and fetch the GraphCompose library.

    Listed21 days ago

    What's inside

    Frequently Asked Questions

    This skill provides a developer-focused harness that uses visual regression and deterministic scripts to convert document references (PDFs/images) into precise, maintainable Java templates, eliminating the trial-and-error of manual coordinate plotting.

    Claude Code and Codex CLI, on Windows, macOS and Linux. It needs Node.js 20+, Java 21+, Maven and ImageMagick locally for rendering and visual diffing; the first-run bootstrap checks them and installs the missing ones only after you agree.

    You receive the complete harness of automation scripts (including bootstrap, rendering, and diffing tools) along with the specialized workflows designed to guide an agent through the document reconstruction lifecycle.

    It produces data-driven templates. The content lives in a JSON data file behind a typed spec, and the layout is built from semantic GraphCompose primitives (sections, rows, weights, anchors) rather than fixed coordinates, so a new invoice or CV is a new data file, and longer content reflows instead of overlapping.

    The skill includes a 'Revision Management' system that tracks changes in a dedicated chain, allowing you to perform ranked visual regression analysis and roll back to previous versions if a layout change fails quality checks.

    Mostly. On first use the agent runs the bundled bootstrap: it checks Java, Maven and ImageMagick (and installs missing ones with your consent), installs the Node dependencies, fetches GraphCompose from Maven Central and registers the workflow skills with Claude Code or Codex. Each run then starts with a preflight against your Java project, and the work is written to a graphcompose-flow/ folder inside it.

    New in Agents & Orchestration