Hysen Labs
Open-source project
programasweights/claudish avatar
programasweights

claudish

Translate between English and Claudish with tiny ProgramAsWeights functions.

215 stars17 forksPythonMIT
DEEP OPEN-SOURCE ANALYSIS

Claudish translates plain English into deliberately over engineered prose

A pair of ProgramAsWeights functions convert everyday sentences into inflated corporate style and back, runnable from Python or a small command line tool.

What Claudish does

Claudish is a small translator between plain English and a style the project calls deliberately over engineered Claudish. The README presents it as two tiny functions built on the ProgramAsWeights platform. One function turns a normal sentence into inflated, corporate sounding prose, and the other reverses the process, turning that prose back into plain language. The project frames this as a parody rather than a serious tool, and it is described as unofficial and not affiliated with Anthropic.

The README gives concrete examples of both directions. To convert plain text, the example passes the sentence Only owners can merge to the English to Claudish function. To go the other way, the example feeds a block of Claudish into the Claudish to English function. That block reads: The honest shape is asymmetric: the data is correct; the format is hard to read. Correctness landed; legibility did not. It is exactly the kind of circular, hedged writing the translator is meant to mimic and undo. A live demo is linked from the README so visitors can try the two functions in a browser without installing anything. The tone of the examples makes the intent clear: the project pokes fun at a certain way of writing by making the transformation mechanical and reversible, which is what separates it from a normal paraphrasing utility.

Using the Python functions

The Python interface is the primary way to call the two functions. The README shows installing the client with pip using an extra index URL that points at the ProgramAsWeights package server. After installation, code imports the library as programasweights and calls paw.function with a program identifier. Two identifiers are given: ca9d5165b6c8e6615529 for English to Claudish and e469f61ccab2699fbd51 for Claudish to English. Calling the function with a string returns the translated text.

The README notes that the programs download once and then run locally, so repeated use does not re fetch them. It also says not to pass a max_tokens argument, because the ProgramAsWeights runtime stops on its own when it reaches the end of sequence marker. This keeps the call simple: hand in a sentence, get a sentence back. The public programs are also published on the ProgramAsWeights hub at the two identifiers above, so the same logic can be used through the website without writing code. The README links the hub pages for both directions. Because the functions are hosted as named programs rather than bundled source, the Python package is thin and the translation logic lives on the platform side, which the README presents as the normal way to use ProgramAsWeights.

Command line and specs

Beyond the Python API, the README describes a small command line wrapper. The example invokes translate.py with a direction and a string, such as translate.py to-claudish followed by a sentence about a release going out after approval, or translate.py to-english with a Claudish style reminder not to launch until tests pass. This wrapper is a convenience layer over the same two ProgramAsWeights functions and lets a user translate text from a terminal without writing Python.

The project also ships specification files that document each direction. The README links specs/english-to-claudish.md and specs/claudish-to-english.md. These specs can be copied into a user's own ProgramAsWeights program or adapted for another model or workflow, which makes the translation behavior portable instead of locked to one account. The README credits gvzdv/claudish-to-english as the inspiration for the project and repeats that it is an unofficial parody not affiliated with Anthropic. There is no build system, test suite, or server component described, and the README does not claim performance or accuracy targets. The scope is intentionally narrow: a pair of hosted functions, a thin Python client, a CLI wrapper, and two spec files that anyone can reuse. That narrow scope is the whole project, and the README keeps the description short to match.

Editorial conclusion

The project is an unofficial parody hosted at github.com/programasweights/claudish and is not affiliated with Anthropic.

DEEP OPEN-SOURCE ANALYSIS

Official sources

Community notes

Community notes