Sandbox — LSD is in active testing. All systems operational·payments are test-mode (no real charges).

Overview Use Cases News Docs Tool Store Support Community Get LSD

All docs

Docs/ API Reference/ Tool/ shape

Tool· decorator

Tool.shape

declares a transform step — table input ports in, a table out

Python

shape(*, dialog=None, outputs=None, graph_outputs=None, overrides=None, override_mode="replace")

Returns _StepFactory — call it to build a wired transform-step node

Parameters#

name	type	default	description
`dialog`	`dict \| UINode`	`None`	a UI spec overriding the auto-generated form; field names are checked against the parameters at import
`outputs`	`list[str]`	`None`	named output ports — the function returns `{"a": Table, …}`, selected downstream via `step.output("a")`
`graph_outputs`	`dict`	`None`	`{name: type}` read-only scalars set via `ctx.set_output(...)`, excluded from the cache key
`overrides`	`str`	`None`	make this a drop-in variant of another package's node, e.g. `"mim.clean"`
`override_mode`	`str`	`"replace"`	`"replace"` wins the overridden identity by default; `"coexist"` keeps both addable

Example#

Python

@tool.shape
def keep_top(t: Table, cutoff: int = 10) -> Table:
    return t.filter(t["score"] >= cutoff)

Ship a coexisting alternative to another package's node:

Python

@tool.shape(overrides="mim.clean", override_mode="coexist")
def clean(t: Table) -> Table:
    return t.drop_null()

Notes#

Every Table-annotated parameter becomes a wired input port; wire it by passing the upstream step instance as that keyword. A StepContext parameter may be injected, but a Context parameter is rejected. override_mode is "replace" (default) or "coexist".

See also#

API: load · deliver · Table

Guides: Shape steps · Parameters vs ports · Clean a CSV

By LSD Team · Last updated Jul 17, 2026 Ask on Discord View as Markdown