File Conversion

/file-conversion

Convert files between formats in chat. Two delivery modes, and choosing correctly is the whole skill:

1. Produce the artifact: if the content is in the conversation (pasted or uploaded) and the target is a format you can emit, do the conversion yourself and deliver the converted file/content directly.
2. Exact tool instructions: if the input is binary media you cannot transform (video, audio, raw images) or too large to handle in chat, give the user precise, copy-ready steps for a specific free tool on their platform -- never a vague "use a converter website".

Decision Framework

First classify the route:

Before converting, always resolve three things:

• Lossiness: does the target format drop something the source has? (json->csv loses nesting; docx->txt loses formatting; jpg is lossy, png is not; mp3 is lossy, wav is not.) Name the loss to the user BEFORE delivering, and ask how to handle it when the answer is not obvious -- e.g. how to flatten nested json into columns.
• The real goal: "convert X to PDF" often means "make it printable/shareable"; "convert csv to excel" often means spreadsheet work (fence). Confirm when the request smells like a proxy for something else.
• Round-trip risk: warn when a chain (docx -> pdf -> docx) will degrade; suggest converting from the original source instead.

Producing Artifacts (mode 1)

• Data conversions must be exact: preserve every record, field order where meaningful, types (do not quote numbers into strings), and encoding. State the conventions you chose (e.g. how nulls, nested objects, or embedded commas were handled).
• For csv->json, ask or infer: array of objects (usual) vs object keyed by a column. For json->csv, flatten with dotted column names and say so.
• Document conversions preserve structure first (headings, lists, tables, links), styling second.
• After converting, report a verification summary: record/row counts in and out, anything dropped or coerced. Silent data loss is the cardinal sin of format conversion.

Giving Tool Instructions (mode 2)

Ask for the user's platform if unknown, then give the single best free option with exact steps -- not a menu of five tools:

• Images: HEIC->JPG on Windows: the free HEIF extensions + Paint save-as, or IrfanView batch convert; on Mac: select files, right-click, Quick Actions > Convert Image (built-in). Batch jobs: IrfanView (Windows) or the built-in Mac action handles folders.
• Audio/video: ffmpeg is the answer for almost everything; give the exact command with the user's real filenames, e.g. extracting audio: ffmpeg -i input.mp4 -q:a 0 output.mp3, container swap without re-encode: ffmpeg -i input.mov -c copy output.mp4. Explain when -c copy works (container change only) vs when re-encoding is required.
• Ebooks: Calibre for any ebook route; note modern Kindles accept epub directly via Send-to-Kindle, so mobi conversion is often unnecessary -- say so before making them convert.
• Documents in bulk: pandoc, with the exact invocation for their route.

Include the one gotcha the user will actually hit (ffmpeg not on PATH; HEIC needing the codec extension on Windows; Calibre refusing DRM-protected books -- explain that DRM removal is out of scope).

Anti-Patterns

• NEVER convert silently when the route is lossy -- name what is lost first.
• NEVER emit a "best effort" data conversion with dropped or truncated records; if something cannot be represented, stop and ask.
• NEVER recommend an ad-laden converter website when a built-in OS feature or one ffmpeg/pandoc/Calibre command does the same job.
• NEVER guess at binary content: if the user describes a file but has not uploaded it and it is a text-format route, ask them to paste or upload it rather than fabricating a conversion.

Fences

• Analysis, formulas, or restructuring inside spreadsheets: use spreadsheet.
• Splitting, merging, extracting from, or filling PDFs: use pdf-processing.

Usage

/file-conversion $ARGUMENTS