nodecode-doc-markdown
Document Markdown: turns one local file — PDF, DOCX, PPTX, XLSX, HTML, CSV, JSON, TXT or MD — into Markdown the model can read. A Nodecode rewrite of TakeshiTGAL/hermes-plugin-doc-markdown by TakeshiTGAL.
Get it
Tell the agent "add the doc-markdown cell", or run nodecode add nc://doc-markdown.
Set it up
Nothing is required: enabled, the tool reads files at once. The doc-markdown section of the config refines it — max_file_bytes (the largest file read, 20971520 by default), max_output_chars (the longest Markdown kept, 200000) and refuse_hidden (refuse a path with a component that starts with ".", on by default). config.example.jsonc is the section written out.
Use it
Ask for the contents of a document and the model calls doc_convert with the file's absolute path: it answers the document as Markdown — headings with their #s, a table row as one line of cells, a slide per ## Slide N, a sheet per ## Name — and says which file, of what type and how many bytes it read. /doc PATH does the same from the shell:
/doc ~/Documents/quarterly.pdf
The guard is the original plugin's, and it is worth knowing before you hit it. Every symbolic link is followed first; then the tool refuses any path with a component that starts with ., any path inside the Nodecode home folder, a relative path, an http, https or file URL, a missing file, a directory, an extension outside pdf, docx, pptx, xlsx, html, htm, csv, json, txt, md, and a file larger than max_file_bytes. What it reads is what this process can open: it is a guard, not a sandbox, and a hostile document is still a document this process parsed.
A .docx, .xlsx or .pptx is read as the zip of OOXML parts it is, with a DEFLATE decoder in the cell, so no Python, MarkItDown, node or npm is involved. A PDF is best-effort: its streams are inflated where they say zlib and the text operators are read out, so a scanned page — or a PDF whose text lives in compressed object streams — answers nothing and says so instead of answering noise. An .xls workbook is not read at all: save it as .xlsx or .csv first.
What is in it
| File | What it is |
|---|---|
cell.lisp |
The section, the path guard, the DEFLATE and zip readers, the OOXML, HTML, CSV and PDF readers, the doc_convert tool, /doc and the manual. |
nodecode-doc-markdown.asd |
The system and its test system. |
test/ |
The offline tests: inflate and zip, every reader over a fixture this process writes, the guard's refusals, the tool through the wire, /doc. |
config.example.jsonc |
The section, written out as documentation. |
License
MIT. The original is Copyright (c) 2026 TakeshiTGAL: see LICENSE and NOTICE.
