atomicbot-ai/atomic-chat/pandoc

pandoc

Convert documents between formats with the universal `pandoc` CLI. Input/output formats are inferred from file extensions; override with -f <from> / -t <to> when needed.

PDF output requires a LaTeX engine (or --pdf-engine). Markdown/HTML/DOCX/EPUB conversions need only pandoc itself.

Setup health check (run first, every session)

Verify with one solo step:

[{ "tool": "os.shell.run", "args": { "cmd": "pandoc", "args": ["--version"] } }]

Outcome map:

• exit 0 + version → ready, proceed.
• command not found: pandoc → enter Setup playbook → "pandoc missing".
• PDF target fails with pdflatex not found → enter Setup playbook → "PDF engine missing".

Setup playbook (when prerequisites are missing)

pandoc missing

Reply (solo reply step):

"pandoc is not installed. I can install it: brew install pandoc. Install it?"

On yes:

[{ "tool": "os.shell.run", "args": { "cmd": "brew", "args": ["install", "pandoc"] } }]

On Linux: apt-get install pandoc.

On Windows, offer this exact winget command after confirmation:

[{ "tool": "os.shell.run", "args": { "cmd": "winget", "args": ["install", "--id", "JohnMacFarlane.Pandoc", "-e"] } }]

PDF engine missing

PDF output needs a TeX engine. Offer the lightweight option:

"PDF export needs a TeX engine. I can install: brew install --cask basictex (or the lightweight brew install tectonic). Which do you prefer?"

tectonic is simplest (pandoc in.md -o out.pdf --pdf-engine=tectonic).

When to use

• "Convert this Markdown to DOCX / PDF / HTML", "turn this HTML into Markdown".
• "Make an EPUB from these chapters", "export notes to a Word doc".

When NOT to use

• Reading a document's text — use os.fs.read_document.
• Editing PDFs structurally (merge/split) — use the pdf skill.
• Spreadsheet conversion — use the xlsx skill.

Common operations

All examples invoke os.shell.run with cmd: "pandoc". Outputs go to the session working directory; the approval gate surfaces each write.

Rules

1. Always specify an explicit -o <output> path and report it to the user.
2. For PDF, prefer --pdf-engine=tectonic if installed (fewer dependencies).
3. Verify the output exists (e.g. os.fs.list) before claiming success.
4. Preserve the source file — pandoc writes a new file, never edit in place.