0rwx:
Do I sometimes blog with AI-assistance? Yup! Is that an offensive waste of time of my fellow human beings? It could be! The YouTuber No Boilerplate published The 0rwx Framework (video), a way of indicating to your readers how much AI was used to produce an article.
So I’m going through all of my articles and remarking whether I wrote them myself entirely, I used
AI for research, but wrote everything myself, I adapted AI-generated material, or the article is
100% YOLO. Or with Markdown metadata: sloppy = true
My motivation is partly to improve quality of my articles: Humans actually read them. Sometimes because I asked them to. Sometimes agents of humans I’ve asked will read them. Very often, they’re prototype workflows that haven’t and might not become skills, but warrants re-enacting at least once.
I only have one blog namespace, and I don’t want to upset my fellow humans who care about artisanal writing (which is still the majority of my articles) or get called out for being a clanker (when I only aspire). So my main motivation for iterating all my articles is to improve quality where I left it below my own standard (i.e. no proofreading or weak towards the end of the article).
Another more sinister goal is to provide a dataset for iterating my blog-post skill using my skill-refinement skill. I’d like to see if I can get really good articles made where the edit distance is shorter. On a good article, I still rewrite 50-60% of it, and what the agent mostly does is code scaffolding.
Every article on this site now carries a small badge in its metadata line that says how much AI was involved. The labels come from 0rwx, a model by Namtao Productions that borrows Unix file permissions:
| Icon | Label | Permission | Meaning |
|---|---|---|---|
hand-made | 0 | AI had no access. I read and wrote everything. | |
ai-inspired | r | AI was read-only. I wrote the work, using AI for ideas and references. | |
ai-collab | w | AI had read-write access. I read and fixed the generated work. | |
ai-made | x | AI made it. I curated output I can no longer fully read. |
Labelling with just ai-usage
¶
I did not label every article by clicking around in an editor.
A small bash script walks through the archive in the terminal:
just ai-usage # only articles without a label
just ai-usage --all # revisit labelled articles too
just ai-usage-report # counts, plus untagged and sloppy articles
For each article it prints the raw Markdown (through bat, less, or cat)
and waits for a single key:
| Key | Action |
|---|---|
1–4 | write ai = 'hand-made', 'ai-inspired', 'ai-collab', or 'ai-made' and continue |
5 | toggle sloppy = true without leaving the article |
s | skip |
r | redraw |
q | quit; the next run resumes with the remaining articles |
The script edits only the first TOML front-matter block, replacing an existing
ai or sloppy line or inserting one before the closing +++. sloppy is a
private note to myself: it is not a tag, so it does not appear in the tag list or
anywhere on the site.