Code
Step · Stable
Run a small sandboxed Python transform over declared inputs to declared outputs.
Execute a short Python snippet inside the workflow sandbox. Declare the context keys the snippet reads via input_ports and the keys it writes back via output_ports. The sandbox provides no network, filesystem, or database access — it is a pure-compute environment for deterministic transforms. Output values are available to every downstream step under the declared output_port names.
When to use
Section titled “When to use”Deterministic transforms that templates cannot express cleanly — e.g. regex extraction, arithmetic bucketing, string slicing, list filtering, or structured data reshaping where an LLM is overkill and a Map step is too coarse-grained.
When not to use
Section titled “When not to use”Anything that needs network access, file I/O, a database, or shared state — use a dedicated node for that. If the transform is a single Jinja expression a Map step is simpler. If the logic is non-deterministic or requires an LLM, use an AI step instead.
Inputs
Section titled “Inputs”Configured per use: code, input_ports, output_ports.
Auto-generated from the skill registry (load_skills()). Do not edit by hand.