Scitex Decorators

Decorator library for SciTeX scientific code. Type-conversion family (`@numpy_fn`, `@torch_fn`, `@pandas_fn`, `@xarray_fn`, `@signal_fn`) lets a function declare its native array type and accept any of the others — the decorator converts on entry and back on exit, including `cuda` round-trips. Caching family (`@cache_disk`, `@cache_disk_async`, `@cache_mem`) memoizes pure functions. `@batch_fn` chunks large inputs through GPU/memory limits; combinator decorators (`batch_numpy_fn`, `batch_torch_fn`, …) compose batching with type conversion. Lifecycle: `@deprecated`, `@not_implemented`, `@timeout`, `@preserve_doc`, `@wrap`. `@AutoOrderDecorator` resolves stacking order so users don't memorise the right decorator ordering. Drop-in replacement for hand-rolled `if isinstance(x, torch.Tensor): x = x.cpu().numpy()` boilerplate, ad-hoc functools.lru_cache that doesn't survive process restart, and per-script for-loop GPU batching.

mkdir -p ~/.claude/skills/scitex-decorators
curl -L https://claudskills.com/skills/scitex-decorators/SKILL.md \
  -o ~/.claude/skills/scitex-decorators/SKILL.md