Inline Python

For the occasional transformation that is awkward to express with Scriptum’s expressions and built-ins, you can drop into Python inline with do ... with python. The code field holds the function body — use a triple-quoted string for readable multi-line code. The function is named handler(event, context); event carries the inputs you pass alongside code, and the return value becomes the step output. (Adapted from examples/python_normalize.scriptum.)

do "normalize text" with python
  code = """
import os
import re

def handler(event, context):
    text  = event.get('text', '')
    strip = event.get('strip_punctuation', False)
    if strip:
        text = re.sub(r'[^\\w\\s]', '', text)
    normalized = text.strip().lower()
    words = [w for w in normalized.split() if w]
    return {
        "normalized": normalized,
        "word_count": len(words),
        "char_count": len(normalized),
    }
"""
  text              = text
  strip_punctuation = strip_punctuation
  env = {
    NORMALIZER_PROFILE: NORMALIZER_PROFILE
  }
  dependencies = ["regex>=2024.5.15"]
  timeout_ms   = 20000
-> result

emit "Done"
  normalized = result.normalized
  word_count = result.word_count

Notes on inline Python:

• Inputs you list (text, strip_punctuation, …) are passed in event.
• env = { ... } injects environment values; timeout_ms bounds runtime.
• A dependencies list declares packages, but dependency resolution is currently limited — prefer the standard library and a small, well-known set of packages, and treat heavier dependency graphs as not yet reliable.

That completes the language tour. For the bigger picture — script, draft, version, thread — see Core Concepts; for the gRPC surface that creates, watches, and resumes threads, see the API Reference.