From markdown_converter
Convert source documents to clean Markdown by delegating to the markitdown-converter agent. Supported sources include PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx), HTML, EPUB, images (.png/.jpg/.jpeg/.webp/.gif/.bmp), audio (.mp3/.wav/.m4a), and data files (.csv/.json/.xml) — single files or whole folders. Trigger this skill WHENEVER the user shares or references such a source and wants to read it, extract it, parse it, dump it, turn it into markdown/text, or make it LLM-friendly — even if they never say the word "markitdown" or "convert". Typical cues include "can you read this PDF", "extract the text from these slides", "parse this spreadsheet", "turn these docs into markdown", "I dragged a folder of reports, summarize them", or simply attaching/mentioning a file of one of the supported types together with any comprehension task. Do NOT attempt the conversion yourself with Read/pdftotext/unzip — always delegate to the `markitdown-converter` agent so caching, image extraction, and semantic image-to-markdown enrichment all work correctly.
How this skill is triggered — by the user, by Claude, or both
Slash command
/markdown_converter:markitdown-convertThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
This is a **dispatcher skill**. Your only job is to recognize that the user's request is a markitdown conversion task, resolve the source path, and hand off to the `markitdown-converter` agent. The agent handles everything else: dependency bootstrap, caching via `convertinfo.json`, image extraction, image-to-markdown enrichment, and folder-tree mirroring.
This is a dispatcher skill. Your only job is to recognize that the user's request is a markitdown conversion task, resolve the source path, and hand off to the markitdown-converter agent. The agent handles everything else: dependency bootstrap, caching via convertinfo.json, image extraction, image-to-markdown enrichment, and folder-tree mirroring.
The trigger is the combination of (a) a supported source and (b) any request that implies the user wants the content in markdown/text form. The request does not need to say "markdown" — "read", "extract", "parse", "summarise", "make it usable", "ingest into Claude", "show me what's inside" all count once a supported source is in play.
Supported extensions (from the agent's folder-mode glob): .pdf, .docx, .pptx, .xlsx, .html, .htm, .epub, .png, .jpg, .jpeg, .webp, .gif, .bmp, .mp3, .wav, .m4a, .csv, .json, .xml. If the source is a folder, trigger if it plausibly contains any of the above — you do not need to enumerate first.
If the user only provides a URL (no local path), do not trigger — tell them to download it locally first, because markitdown-converter expects an absolute filesystem path.
Resolve to an absolute path before calling the agent. Expand ~, resolve relative paths against the current working directory, and on Windows pass a normal path (the agent normalises slashes internally). Do not quote-wrap the path inside the prompt — just include it as text.
Decide the output folder.
./out"), pass it explicitly.out and let the agent default to <cwd>/tempMarkdown. Do not pre-create the folder yourself.Invoke the agent via the Task tool with subagent_type: "markitdown-converter". Use a single, focused prompt that names the source (and optional out) and otherwise trusts the agent. Example:
Convert this source to markdown:
source: C:/Users/c0917/Downloads/2025-annual-report.pdf
Or with a custom output and a folder:
Convert this folder to markdown:
source: C:/Users/c0917/Documents/client-docs
out: C:/Users/c0917/Documents/client-docs-md
Relay the agent's summary back to the user — total files converted vs. cached, output folder, and convertinfo.json path. Don't reformat or re-narrate it; the agent's summary is the answer.
Read and paste contents. Binary formats (PDF, docx, pptx, xlsx, images, audio) cannot be meaningfully read that way, and for the text formats you'd be duplicating work the agent caches.markitdown, pdftotext, unzip, pymupdf, or any extraction yourself. The agent auto-installs its deps and caches by sha256 so repeated conversions of the same file are free — bypassing it loses that.tempMarkdown/ or assets/ folders. The agent owns that tree.convertinfo.json cache relies on one run seeing all files..rtf, .odt): tell the user markitdown may or may not handle it; offer to try anyway by passing the path through, and warn that the agent will skip unsupported files in folder mode..md file.Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub syunhsieh/claude-marketplace --plugin markdown_converter