alchemy-writing 0.1.0

Turn AI prose into human-quality writing. A drop-in ruleset that stops your coding agent from writing like an LLM.

alchemy-writing

Turn AI prose into writing that reads like a person wrote it. This is the Python distribution of Alchemy — a small ruleset you hand your coding agent so the docs, READMEs, and replies it writes stop carrying the tell-tale LLM patterns.

Install

pip install alchemy-writing
alchemy init

alchemy init writes ALCHEMY.md into your project and links it from your CLAUDE.md, AGENTS.md, or .cursorrules if one exists. Your agent reads the rules whenever it writes after that.

Print the rules instead of writing them:

alchemy print

Or pull them into your own code:

import alchemy_writing
print(alchemy_writing.rules())

What's in it

The rules are grounded in documented AI-writing tells: Wikipedia's "Signs of AI writing," the Kobak et al. study on words that spiked in research papers after ChatGPT, and Pangram's phrase-frequency data. They cover the banned constructions, the fingerprint vocabulary, the punctuation tells led by em-dash overuse, and a self-check the agent runs over its own prose.

Full project, including the npm package and MCP server: https://github.com/fernforge/alchemy

License

MIT