Copy this first
/pstack:technical-writing Write a README for this Node CLI. First tell me which doc type each section should be, and write it once I've confirmed.AI-written READMEs often have a bit of everything: install steps, philosophy, and an option table, all mixed up. Beginners and people looking up options both have to skip around.
The first rule of /pstack:technical-writing: one doc does one job. It sorts each section into one of four types:
- Tutorial: follow along once, and see a result at every step.
- How-to guide: steps to solve one specific problem.
- Explanation: why it’s designed this way, and the tradeoffs.
- Reference: options, flags, error codes. Facts only.
If a README needs several types, split it into parts that link to each other.
A Node CLI example
A CLI that batch-exports Markdown notes to HTML needs a README for GitHub:
/pstack:technical-writing Write a README for this CLI.
Write install and the first export as a tutorial: at each step, tell the reader what output they should see.
Put all options in a separate reference section, built from the real output of --help. Don't make up any options.
When you're done, follow the tutorial from scratch in an empty folder, make sure every command works, and paste the terminal output.
Finally, run /pstack:unslop over it: no em dashes, no "powerful," no "effortless."- “Follow the tutorial in an empty folder”: the most useful line. You can argue about style. You can’t argue about whether a command runs.
- “Built from the real output of –help”: reference docs state facts, and the facts come from the code.
- “Finally, run /pstack:unslop”: strips the AI tone, plus any words you don’t want.
Where to look after you send it
At the start: how it sorted the types. If philosophy ended up in the getting-started part, correct it right away.
At the end: the terminal output. If a step fails, the tutorial is missing a step.
Strip the AI tone at the end: /pstack:unslop
/pstack:unslop the readme changes, no emdashesIts built-in rules target common AI tone in English: overused em dashes, forced lists of three, empty endings. Add any words you care about to the rules, like “no ‘seamless,’ no ‘empower.’”
Code others will call: write the tutorial first
For a package or API that other code will use, you can write the docs before the code:
/pstack:poteto-mode write a tutorial for how i would use the new config package first. then /pstack:teach me why it beats the current one.Writing the tutorial first makes it think like the user: what to write first, and what you’ll see. The tutorial also becomes the target to check the code against.
Not sure you asked the right thing: /pstack:sharpen
/pstack:sharpen Write onboarding docs for the new billing service.It doesn’t write the docs. It rewrites your request to be sharper. For example: what an expert would also do, what source material they’d start from, and what could prove the result right. Read the rewrite. If it looks right, send it.
Stuck?
The README still has a bit of everything.
Reply: “Split it the technical-writing way: keep only the tutorial in getting started, move the option table to the reference section, and give the design ideas their own explanation section. Don’t mix them.”
I ran /pstack:unslop, and it still sounds like AI.
List the exact words and phrases that bother you, and add them after the command. For example: “/pstack:unslop this section, no em dashes, no ‘seamless’ or ‘empower,’ sentences under 20 words.” The more specific the rules, the better it does.
/pstack:sharpen didn't write my docs.
It only rewrites the request. Send the rewritten request to /pstack:poteto-mode or /pstack:technical-writing.
The PR description is too long.
Reply: “A PR description is a one-minute brief. One to three sentences per section. Put detailed data in a linked file.”