Convert To Markdown

Convert To Markdown

Use this skill before reading local documents that are likely to be expensive or awkward in their native format: PDF, DOCX, PPTX, XLSX, HTML, CSV/TSV, JSON, XML, or similar reference files.

The goal is simple: create a clean Markdown version first, then read the Markdown instead of dumping raw binary/native document content into context.

When To Use

• The user points to a local PDF, Word document, PowerPoint, spreadsheet, HTML export, CSV, or data file.
• ovd-workflow knowledge vault is ingesting non-Markdown references from .overdrive/knowledge/.
• The task is summarization, extraction, comparison, source review, requirements analysis, or context preparation.
• A model/native file upload would consume unnecessary tokens or obscure tables/headings.

Do not use this skill for tiny plain-text files, source code, images that need visual inspection, or files where exact layout fidelity matters more than text/structure.

Preferred Tool

Use Microsoft MarkItDown when available:

markitdown "/path/to/source.pdf" -o "/path/to/source.md"

If MarkItDown is missing and the user wants setup help:

python3 -m pip install 'markitdown[all]'

Use a project virtual environment when possible. The base package is lighter, but [all] is the practical install for PDF/Office/spreadsheet coverage.

Workflow

1. Identify the source file and pick a safe output path beside it or under a local cache folder.
2. Check whether markitdown exists:

command -v markitdown

3. Convert with markitdown input -o output.
4. Read the Markdown output, not the raw original, unless conversion failed or layout/visual fidelity is the point.
5. Report conversion caveats briefly: missing tables, OCR limits, unsupported format, or skipped images.
6. Never overwrite an original file. If an output path exists, use a cache path or ask before replacing it.

ovd-workflow Knowledge Vault

For .overdrive/knowledge/, prefer the runtime command:

overdrive knowledge --apply

This refreshes .overdrive/knowledge-index.json, converts supported non-Markdown files to cached Markdown where possible, and leaves summaries/topics for the agent to fill in later.

When using the vault during a task, inspect knowledge-index.json first, then load only the relevant original file or markdownCache. Do not dump the full vault into context.

Fallbacks

• If MarkItDown is unavailable, ask whether to install it or read the file natively.
• For simple .txt, .csv, .json, .html, .xml, or log files, a plain fenced Markdown cache is acceptable.
• For web pages, use defuddle or normal browser/web-fetch workflows instead of forcing MarkItDown.
• If a PDF is scanned/image-only, note that OCR may be needed and do not pretend conversion captured all content.

Honesty Boundary

Overdrive cannot intercept every native file ingestion path inside Claude, Codex, Cursor, Gemini, or Antigravity. This skill is instruction-led: convert local files before reading when possible. ovd-workflow also uses the same idea during knowledge-vault ingest.

Security

MarkItDown runs with the permissions of the current process. Convert only files the user provided or that are already in the project/workspace, avoid untrusted remote URLs, and never expose secrets found in converted documents.

Attribution

Built around Microsoft's MarkItDown project: https://github.com/microsoft/markitdown