Scriptty

Passive workflow documents over explicit Python pipelines. Python 3.11+.

from scriptty import load
workflow = load("examples/text.json")
metadata = workflow.inspect()
bound = workflow.bind(selected_catalog, catalog_version="text-v1")
result = bound.run({"text": " Hello WORLD "})

The host supplies a Hidden Moves CapabilityCatalog. Parsing never discovers or imports providers, evaluates code or invokes operations. JSON and TOML share the same strict schema-1 model; see examples. A document contains schema, name, version, steps and outputs. Steps contain id, capability and arguments. Expressions carry exactly one tag:

Paths contain string keys or nonnegative integer indexes. All IDs, shapes, references and document keys/versions are checked passively. Binding validates selection and known literal argument shapes. Run preflight validates known input values; producer-dependent paths/values are checked before their consumer runs. Errors include source and structural step/expression location; JSON syntax errors include line/column. TOML syntax diagnostics come from the standard library.

Documents are bounded to 1 MiB and JSON depth 32, with at most 256 steps and paths of at most 32 segments. Duplicate keys, non-finite values, invalid UTF-8, unknown tags/keys/versions and TOML dates are rejected. No new grammar is introduced. Schema 1 is stable for this source line: incompatible definition changes require a new schema number; unknown fields are rejected instead of silently changing meaning. No terminal, stdin or signal behavior is installed by library imports.

The independently versioned source profile pins moveset and its selected siblings. Build with uv build --out-dir dist; python scripts/qualify.py dist qualifies both installed artifacts outside the source tree. Publication and released dependency requirements are separate Project tasks.