wolfram-headless

# wolfram-headless — reliable headless Wolfram in Claude Code Hard-won rules for running `wolframscript` from Claude Code's Bash tool. Two findings dominate: (1) **"The product exited because of a license error" is almost never a license problem** — it is a kernel **crash** (usually a memory spike) mis-reported; and (2) **literal Greek Unicode in `.wls` files silently corrupts symbols** under wolframscript. Both cost real time; this skill encodes the fix. ## RULE 1 — ASCII escapes in `.wls`, NEVER literal Greek (silent-corruption bug) `wolframscript` reads `.wls` files under a non-UTF-8 encoding, so a literal `ω` (U+03C9) becomes a **different, dead symbol** (mojibake `ω`), with **no error**. `ω3 =!= \[Omega]3`; any `===`/pattern test against the real symbol silently fails and results are wrong. Verified: a function keyed on the real `\[Omega]3` falls through / returns garbage when fed the literal-`ω` symbol. - In every `.wls`, write `\[Omega] \[Alpha] \[CapitalLambda] \[CapitalDelta] \[Pi] \[Epsilon] \[Tau] \[Mu] \[Xi]`, never the literal characters. `Pi`, `I`, `Infinity`, and ASCII names (`Delta`, `Esq`, `tau`) are fine. - The notebook/desktop kernel (Mathematica FE / Wolfbook in VS Code) handles literal Greek fine — this is **headless-`.wls`-only**. So copying notebook code into a `.wls` introduces the bug. - Quick converter (run before any heavy job): replace the Greek chars with their `\[...]` escapes. A one-liner is in `scripts/greek2esc.py` next to this skil...