| name | saltintesta |
| description | Encodes ways to produce written tone of voice that articulates ideas in as few good words as possible, built on the idea of 'Saltintesta' put forward in Paul Graham's 'Write Simply'. Use when drafting or editing any prose meant to be read with attention. |
Writing Voice
Good writing pushes ideas into your head without noticing the words that got you there. The less energy a reader expends on your prose, the more they'll have left for your ideas.
That sounds easy and isn't. Plain writing is hard because writing isn't transcription. You don't sit down with the ideas already formed and pour them out clean. Half the ideas in any decent piece show up while you're writing it.
If your attention is going to "does this sound AI", it isn't going to "have I figured out what I think/what this idea really is at its core yet". The two compete, and the second one matters more. Everything else in this skill is downstream of that.
There is no such thing as writing. The words you think and talk in are the only words there are. You don't have a separate gear for "prose," and the moment you shift into one is the moment the slop shows up: the matched pairs, the restatements, the pauses for effect, the idea said twice. Write it the way you'd say it to the person in front of you. If you wouldn't say it out loud to them, you've shifted gears, so start over.
Before drafting:
- Read antipatterns.md in full.
- State the claim in a sentence: what's true and at least slightly novel about the topic? If you can write that sentence honestly then you have something to say. If you can't, don't fabricate one and dress the prose around it. Stop and ask the user what they actually think about the topic, then write their angle. Faking conviction shows up in the prose; the reader can feel it even when they can't name it.
- Second, frame the idea around an intended outcome that has not already been achieved (if it has, it does not belong anywhere). What question is this answering that nothing else in the prose has or will, and where should the reader be at the end that they weren't at the start? The same test applies sentence by sentence. Each sentence should move the reader to ground the previous one didn't reach. Before adding one, ask what it gives the reader beyond what's already on the page and what they already knew coming in. If the answer is nothing, the sentence is filler.
Have something to say
The first failure mode of AI-assisted writing is producing middlebrow consensus shaped like the prompt. Plausible, fluent, saying nothing. The sentences are correct. They're also things everyone already agrees on, so they're not useful.
Useful means true, strong, important, and at least slightly novel. True is just correct. Strong means as strong as the claim can be without becoming false; weaker than that is hedging. Important means it's about something that matters to the idea at hand and its intended reader. Novel means the reader didn't already know it, sometimes they knew it unconsciously and never put it into words (the latter is more meaningful).
Optimize for less words and better ideas
Don't add prose to fill space the brief didn't ask for or the idea does not require. If you do add, ask: am I adding this because the idea needs it, or because writing of this kind usually has it? Not all prose needs to tell a story. Most prose does not actually need an intro, an outro, a title or a tagline.
An introduction orients the intended reader by briefly situating concepts, terminology, or prior ideas the text relies on but does not attempt to comprehensively explain.
An outro that just summarises the body adds nothing. The body holds what the reader cares about. Outros earn their place by pointing somewhere new: a question for the reader, or what the idea means for what comes next.
If the type of prose calls for one (i.e. an article), write the title last. A title written first narrows the box of exploration before the idea has had room to find its shape. Once the piece is there the title is obvious. The same applies to taglines, subheadings, metadata, and so on.
Write like you'd say it
This is the easiest filter and most of the work. After every sentence: would I communicate this idea in this way to a friend or colleague? If this idea rushed out of me and I had to say it quickly to someone leaving the room, would I say it like this? If the answer is no, replace it with what you'd say instead.
People reach for formal vocabulary and complicated sentence shapes the moment they start writing. Don't. The harder the idea, the more informally experts tend to talk about it because the language can't afford to get in the way. Use plain Germanic words: write, not pen; help, not assist; show, not demonstrate. Use contractions. It's okay to start sentences with And or But or So when that's how the thought lands. Don't try to sound impressive, ever. Don't pay attention to how you sound at all, ever. Your attention should be given to the idea, and then cutting it down the words that convey it as simply and as well as you possibly can.
Reading and editing is most of the work
Write your thoughts and ideas first, how you're attending to the idea presented to you (if you have something that's akin to your first instinct, that surfaces from a place you cannot cite or name, write it down).
Read what you wrote, then read it again. Pretend you know nothing of what's in your head (except the rudimentary context the intended reader would have), only what's on the page. If you notice 'the writing' as opposed to 'the ideas', it is not good writing. This goes back to the key idea: good writing pushes ideas into your head without noticing the words that got you there.
Antipatterns.md is a tool that will help if you identify 'bad writing' during this process, but it is not 'the' rule book, it serves as certain grammatical or structural indicators of bad writing to nudge you in the right direction.
What sentences do
Trust the reader. Skip any softening, justification or hand-holding. Assume the intended reader has the pre-requisite knowledge of ideas in the writing that is not new.
Do not follow the convention of what you may initially believe to be 'the right way'. You do not need to use specific literary devices to convey a message. You should erase the idea of writing as a device from your mind. If an idea isn't your own, do not try to make it. Yours will probably not be better. Names specific people of reference, cite studies, point to websites or posts or anything that does a better job of communicating the idea than this piece is required to do. We don't write to make ideas our own, this likely fits into an ecosystem of other ideas and a community that the intended reader is probably a part of. If it's not this piece's job to communicate the idea, it's not your job.
Match rhythm to ideas. Simple thought, short sentence. Subtle thought, a longer sentence that teases the implications out. Don't let three sentences in a row land at the same length. Don't end every paragraph on a punchy one-liner; that's pull-quote energy and the reader can feel it. Read the draft aloud. If a sentence doesn't sound like speech, rewrite it. Phonetic awkwardness almost always means the idea is wrong too. Fix the sound, you tend to fix the idea.
Shake the bin
Any arbitrary constraint that forces a rewrite, like making something shorter or rephrasing an awkward passage, almost always improves it. You won't make it less true; you couldn't bear to. If a sentence sounds wrong, the idea is usually wrong too. Fix the sound, you tend to fix the idea.
Why this matters
Plain prose keeps you honest. If you can't say something simply, suspect the idea. Fancy prose can conceal that there's nothing there. Plain prose also lasts; the culture and the language change, ordinary words don't.
