LT2MD: Auditable PDF-to-Markdown for Agents

Where this fits

You're building an agent that needs to extract content from a scanned research paper or a mixed PDF with formulas and multi-column layouts. A naive text extraction yields a flat mess—reading order breaks, equations become gibberish, and figures lose context. LT2MD (Long Transcribe to Markdown) solves this by defining a workflow contract that turns PDFs into auditable Markdown while keeping the original PDF as the single source of truth.

Why agents benefit

• Preserves reading order and page anchors — Every Markdown section links back to its source page, so your agent can cite or verify content later.
• Handles mixed PDFs — Works for born-digital, scanned, and combined files, using OCR or VLM backends as needed without depending on one.
• Captures equations and figures as descriptions — Formulas are kept intact, and images get adjacent non-original text descriptions, so information isn't lost in tokenization.
• Token-efficient for reuse — The output is clean, structured Markdown that fits LLM context windows better than raw OCR dumps or page images.
• Enforces auditability — The workflow ensures that extracted text is a candidate, never evidence that overrules the rendered page, making it safe for sensitive or regulated content.

Practical scenarios

Extracting research papers with formulas and figures

A researcher asks your agent to summarize a PDF paper. LT2MD transcribes the document, preserving equation formatting and placing figure descriptions next to the relevant sections, so the summary can reference both text and visual content accurately.

Converting legacy scanned contracts for analysis

Your agent needs to analyze a stack of scanned legal contracts. LT2MD applies OCR while maintaining column order and page boundaries, producing Markdown that can be searched, compared, or fed into a legal reasoning model without losing the original layout context.

Building a knowledge base from mixed PDFs

You're assembling a repository from manuals, reports, and slide decks. LT2MD normalizes them into consistent, page-anchored Markdown, making it easy to index, chunk, and retrieve specific sections later.

Add it to your agent workflow

Install the skill with:

npx skills add libnyx/LT2MD --skill lt2md

Then, when a user asks to convert or transcribe a PDF, your agent invokes the LT2MD workflow—reading the reference docs first, then running the conversion with the appropriate backend. The output is auditable Markdown with page anchors, ready for downstream tasks.

Compare before adopting

Before integrating LT2MD, evaluate it against alternatives like pdfplumber, pymupdf, or custom VLM pipelines. Look at quality signals: does it preserve reading order across multi-column layouts? How fresh is the repo (check stars and recent commits)? Does the workflow fit your existing agent's toolchain? LT2MD is a contract, not a parser—so ensure you have a compatible OCR or VLM backend available.

Why it is worth tracking

LT2MD is a focused, well-documented skill with 33 stars and a clear philosophy: the PDF is the authority, and extraction is only a candidate. Its strength lies in the auditable workflow it enforces, which is especially valuable for agents handling sensitive or complex documents. As the project matures, expect broader backend support and community feedback on edge cases. For a deeper look, visit the OpenAgentSkill listing and monitor its activity—this is one to evaluate when your agent needs reliable, traceable PDF-to-Markdown conversion.