2026-07-06 Mon

Filed
Updated

2026-07-04 Sat

Ideas
- Banning AI from using specific words or phrases
  - I'd like to make a system that makes it very easy for me to ban AI from using words and phrases with me or when creating documents.
    - My first thought about this was to make an AI/Agent/Skill in rulesync which would let me quickly add a word or phrase to a part of the knowledge garden that catalogs the words or phrases that I'd like AI to avoid when communicating with me or on my behalf, following progressive disclosure. I was thinking that ideally, there would be a table of contents page that would have logseq wikilinks to individual pages, one per word or phrase. That way listing the contents of a namespace would be sufficient to tell the AI what to ignore.
    - Later, I was thinking of using Claude/Code/Hooks for this.
- Convert #tag links to [[wikilink]] form on commit
  - I'd like a pre-commit converter, built like the text:fix-invisible mise task, that rewrites Logseq #tag and #[[Page]] links into [[...]] form — one obvious way to link (Py/Zen of Python), and the shape Looksyk can import, since it reads [[...]] but not #tag.
  - Not as safe a rewrite as invisible whitespace, so it stays deferred. A # also begins headings, hex colors, shell comments, and URL fragments, so the converter has to skip fenced and inline code, ignore # before a space or another #, and treat # as a link only at a word boundary. Scope it to .md files under pages/ and journals/, where well over a thousand inline # links live today.
Filed
- fail2ban
- Person/James Clear
- Person/James Clear/Book/18/Atomic Habits
- Person/James Clear/Book/18/Atomic Habits/Quote/Good Easy, Bad Difficult
- My/Principle/Make the Right Thing Easy and the Wrong Thing Hard
  - Intent (your note integrated)
    Core philosophy: make the right thing easy and the wrong thing ridiculously hard.
    
    → Golden paths = paved, safe, fast.
    
    → Off-path = friction, reviews, warnings, incompatibility, or simply impossible.
- Py/Zen of Python
- My/Principle/Simplify/Not every distinction is worth recording
- Terminal/Multiplexer/Idea
- Logseq/Entity/Dotfiles/Dep/Idea
Updated
- dependency intake vision
  - My/Dotfiles/Process/Adding a New Global Dependency/Idea
- light formatting
- mixed-metaphor example
  - My/Pref/Writing/Avoid Distractors such as Awkward or Superfluous Metaphors
  - Word/Salad
- words to avoid
  - My/AI/Rule/How to Communicate Effectively With Me/Words and Phrases to Avoid
  - Logseq/Entity/Software/Project

Linked references — 5 pages

Open the linked references

2026-07-03 Fri

so much Word/Salad

ugh. AI keeps writing things into my plans to "never" do something that's completely unrelated, e.g. while building an iPhone app, "never talk to a brown cow" (it's usually not that unrelated - but that's how it feels). So often, AI expressions collapse into an always/never binary.
so many double negatives.
That's why example 1 (wrap/linebreak) felt like a non-example to you — an unfamiliar reader doesn't obviously know what linebreak does, so spelling it out is earning its place under the real preference, even though the spec's current wording would tell an agent to omit it.

regarding My/AI/Rule/No Recipe in the Cake - new My/AI/Rule/Prune useless commandments

❯ in what universe does this make sense?

  > ### Requirement: Link rationale is not narrated into the config

  > A config comment carries the garden link. It SHALL NOT explain the linking rule itself
  — for example, why the link points at an entity page rather than a preference page.
  Comments can still say what a file or setting is; they just don't explain the linking
  convention.


  to me, this is just terrible. It's so terrible I can't even really articulate what
  preference I should add to my knowledge garden so as to guide AI away from this. I keep
  noticing that AI puts into rules explicit instructions not to do things ... that it
  should be obvious one should never do. Imagine a parenting guidebook that said, "when
  teaching your child how to tie their shoes, you must never push them into the water."
  the instruction in the 2nd part of the phrase doesn't have anything to do with the
  first part of the phrase. In what world would it make sense to make a comment that
  itself explains how and why the comment is being made?

⏺ You've named the actual disease, not just a symptom. This requirement only exists
  because AI narrates its own instruction-following into things it produces. But "don't
  narrate why you added a link" isn't a fact about how garden links work — it's a symptom
  of that one general AI failure mode, bolted onto this one spec as if it were a
  link-formatting rule. Your shoe-tying analogy is exactly right: the two halves of the
  sentence have no real relationship to each other except that AI needed reminding not to
  misbehave, and someone decided the reminder belonged wherever the misbehavior happened
  to show up last.

  The general fix already exists, once, in the one place it belongs: the "No Recipe in the
  Cake" AI rule, loaded into every session via .claude/rules/. A domain spec about how
  links work shouldn't also carry a copy of "and don't narrate yourself while doing this"
  — that's not a property of links, it's a standing constraint on AI's writing that
  applies everywhere, all the time, already.

  I think the right move is to delete this requirement entirely from garden-entity-links

Filed
Updated
- added a line on flowery synonyms
  - My/Pref/Writing/Use the simpler word
- added new Neuralese examples
  - Word/Salad
- clarified the explain-arcane-settings guidance
  - My/Pref/Dev/Management/Configuration/Comment Style
- documented a same-day violation of the rule
  - My/AI/Rule/No Recipe in the Cake
- noted the experimental skills-vendoring technique
  - My/Pref/Dev/AI/Config/Management/Rulesync
- renamed Poetry to PyPoetry
  - PyPoetry/Concept/How to handle Monorepos in Poetry vs uv
  - PyPoetry/Concept/Misconception - Disadvantages of Pip vs Poetry
- renamed and rewrote the CLI setup tutorial for the official readwise cli
  - Readwise/CLI/Tutorial/Install and Log In
- repointed Scarvy's rw-cli link after the tutorial rename
  - Person/Scott Carvalho
- updated with deptrust
  - Security/DevSecOps/Taxonomy
- updated with official readwise cli
  - Readwise/CLI