ware

penkki

A bench for instruction files: runs every case through the real model and leaves the receipt.

since

October 2026

stack

C# · .NET 10 · podman

state

private

what it is

A document for a coding agent is a program with no compiler. penkki is its test harness. A bench is a directory that declares conditions, prompts and runs. penkki run puts every cell through the real claude, each reply in a fresh home inside a rootless podman container of its own, and keeps what came back: the diff, the readings before and after, the receipt.

what it does

• Answers whether a document does its job on a named reader, with numbers.
• Every library on this site was cut from a document and checked on this bench before a human read the code.