Execute a saved trusted Python task without opening an authoring workbench. The runner writes package modules and a workflow into a fresh run directory, launches a separate interpreter, then retains the request, result and source hashes.
Python 3.11 or newer.
python -m pip install -e .
python -m examples.demoThe complete task contains a formatter module plus run_workflow; the payload contains {"name":"avery"}. The example proves that an unauthorized call is refused, then runs the authorized task and returns {"greeting":"Hello Avery"}. Temporary runtime files are removed; the printed evidence path remains.
Separating task materialization from the authoring workbench makes saved work independently executable and reviewable. Read the mechanism and implementation notes for the specific boundaries and source links.
For the failure path, run python -m examples.failure_walkthrough; the
authoring and evidence guide explains what survives.
This executes arbitrary trusted Python. A fresh process is not a filesystem/network security boundary, and timeout handling does not establish isolation of descendant processes. There is no model connector or dependency solver.
python -m pytest runs the behavior tests (install pytest first). The runnable example above provides a separate first-use check.
MIT licensed; see LICENSE.md. Origin and release boundaries are documented in ORIGIN.md and SECURITY.md.
Open the saved synthetic result alongside its input and demonstration. The result is from the bundled synthetic example; local machine paths and temporary run identifiers are excluded from public projections.