nb-to-wolfbook

nb-to-wolfbook

Converts Mathematica files (.nb / .m) to Wolfbook .wb (a VS Code Notebook JSON), and — critically — makes the resulting cells bridge-safe so they evaluate identically through the Wolfbook MCP (runCell / evaluateExpression) as they do in the Mathematica front end.

When to invoke

/nb-to-wolfbook <file-or-directory> — convert .nb/.m to .wb (bridge-safe + PUA→ASCII). /nb-to-wolfbook --fix-wb <file.wb> — normalize an EXISTING .wb in place (bridge-safe + PUA→ASCII), backup written. /nb-to-wolfbook --puafix <file.wb> — de-rectangle ONLY (PUA→ASCII, no newline reflow): the smallest diff that removes the rectangle glyphs from an already-bridge-safe .wb.

⚠️ The bug this skill exists to prevent (READ THIS)

A Wolfram cell can wrap one statement over several physical lines:

Eeqv[{j1_},{k1_}] := (-1)^(j1) j1!/(k1-1) Coefficient[
   GenSer, ep[j1,k1]];
MMVbar[j_,k_] := (-1)^(Total[j]+Length[j])
   MMV[j,k];

The front end parses each cell as one unit, so this is fine interactively. But an evaluator that splits a cell on newlines (the Wolfbook MCP runCell, a headless Get of re-serialised text, naive ToExpression) sees broken fragments and either:

• throws Syntax::sntxi: Incomplete expression (the Sum[...,⏎{...}] case), or
• treats the newline as implicit Times: (-1)^k ⏎ MMV[...] silently drops to a product / loses the MMV factor, and def1;⏎def2; becomes Times[def1;, def2;] instead of two definitions.

This corrupts the kernel silently — definitions look present but are wrong — and is exactly what makes a cell-by-cell rebuild "unfaithful." The fix: put each statement on one physical line, collapsing only the intra-statement newlines (insignificant whitespace) and keeping a newline only after a top-level ;. Semantics are unchanged, comments are preserved, and the cell is now safe for any evaluator.

The subtle case that slipped through once (the MMVbar regression)

MMVbar[...] := (-1)^(Total[j]+Length[j]) MMV[...] is one RowBox (the space is implicit Times), merely soft-wrapped for display in the .nb. The FE's "InputText" export ignores PageWidth (verified: Cell option, ExportPacket option, wrapping-Notebook option — all still wrap) and turns the wrap into a hard newline with a continuation indent: ... Length[j])\n MMV[...]. The line ends in ) — a complete value — so the old "does the line end in an operator?" heuristic said "statement boundary, keep the newline," and the MMV factor was dropped (MMVbar became a bare sign). This sat latent in the committed .wb/.nb for months.