chore: introducing the Mycelium AI Policy (#566) · hawkw/mycelium@b77e854

system"! please bear in mind that this is a hobby project that i work on in my

spare time, so pull requests may not always be reviewed in a timely manner.

## AI policy

when contributing to Mycelium, an alleged "operating system", you are free to

use any tools you deem appropriate. this includes the LLM of your choosing,

agents, loops, gastown, AI code review tools, Amazon Rufus, or even `vim`.

however, all contributions to this project (including my own) must abide by

exactly one rule:

**your code (and any associated prose, such as commit messages and

documentation) must not _look_ like it was written by a large language model.**

basically, the game is simple: contributors are free to use whatever AI tools

they wish when contributing to mycelium and associated projects, but in order to

have your PRs merged, you have to be able to *trick me* into believing it was

written by a human being. any PRs which appear to have been generated by a model

will be closed as a violation of the Mycelium AI Policy (even if they were

actually written on pencil and paper by a human being in a cave with no access

to the internet). repeated violations of the Mycelium AI Policy may result in a

ban from contributing to the repo.

### advice on AI policy compliance

+ when using an AI coding agent, the easiest way to ensure your contributions

are compliant with the Mycelium AI Policy is to read the code that the model

generates, and rewrite any parts of it that "feels like Claude-babble". repeat

this process as needed until it no longer feels like it was written by a

model.

+ you may feel that it is possible to prompt an LLM into outputting code that

doesn't look model-generated. my personal experience has been that this is

impossible, and model-written code always requires some degree of rewriting to

make it look Normal. if you think you're better at prompting than me, you are

welcome to try it, but...good luck!

+ while contributors are free to disclose the use of AI tools in their pull

requests, note that this will put you at a substantial disadvantage, because

an AI disclosure will mean I go in suspicious. if you can trick me into

believing that a pull request which says "This code was written using Claude"

was written entirely by an organic, free-range human being...well, I'll be

impressed.

### rationale

the Mycelium AI Policy may, at first blush, seem a bit strange: why am I

explicitly *encouraging* potential contributors to attempt to mislead a

maintainer? and, doesn't the "you have to trick me" policy seem a bit...

adversarial? to answer some of these questions, allow me to share some of the

thinking behind the Mycelium AI Policy, and why I believe it is the best way to

achieve *my personal goals* for this project.

1. this project is a labor of love. I am doing this purely for fun, in my spare

time. reviewing thousands of lines of slop is not particularly fun. reading

the bizarre blend of Rationalist slang, technical jargon, and nonsensical

metaphors that Claude's comments always seem to regress into is particularly

infuriating and I don't ever want to see it in my weekend project.

3. I believe that code can be beautiful. I love code which is both technically

effective *and* a form of creative self-expression. I want the code in

Mycelium to be beautiful. and I have never found purely model-generated code

to be beautiful.

4. however, I have also found LLMs to be an effective tool for generating large

amounts of code faster than a human being can type it. i have also found them

to be useful tools in applications where the model is *not* writing the code:

while an LLM review may not question the high-level design of a feature the

way a human reviewer would, they *are* remarkably thorough, and good at

finding the one place where you forgot to change something or didn't use

checked arithmetic or whatever. i don't want to completely forbid the use of

a tool which i *have* found useful in some particular, restricted ways.

5. i don't want to merge code that is not well-understood by a human being. i do

not believe that it is possible for an engineer to take responsibility for

code that they do not understand thoroughly. in LLM-assisted programming,

there is an obvious temptation to not read the code that the model generates.

we can all fall victim to this temptation from time to time. in my own work,

i have found that the act of rewriting the model-generated code to not Look

Like That forces the person who generated it to actually read and understand

it thoroughly.

6. it's my project and I can do whatever the hell I want to.

## building & running

mycelium is built using [`inoculate`], the horrible mycelium build tool.

[`direnv`]: https://direnv.net/

[`cargo nextest`]: https://nextest.sh

[`justfile`]: #justfile

[Just]: https://just.systems

[Just]: https://just.systems