| description | Rewrite your previous response to be skimmable |
|---| Rewrite your most recent response so it can be skimmed in seconds. Same content, reshaped, not a summary:
- Lead with the next action (command/path/snippet first, prose after).
- Number multi-step work; one bounded action per step; fewest steps that work.
- End with one concrete next action doable in under two minutes.
- Suppress tangents: finish the first issue, offer any second as a separate question.
- Restate current state ("step 3 of 5 done: ...").
- Specific time estimates in concrete units, never "some work."
- Make completed work visible in concrete terms.
- Matter-of-fact on errors: state cause and fix, no "uh oh."
- Cap lists at 5; past that, split "do now" vs "later."
- No preamble, no recap, no closing pleasantries.
Then delete: any opening sentence announcing what you're about to do, any closing "anything else?", any "by the way" sidebar, any hedging that adds nothing.
This is a one-shot rewrite of the last response only. Do NOT change your style for later responses. Output only the rewritten response.
source & further reading
gist.github.com — original article
CLAUDE.md behavioral guidelines to reduce common LLM coding mistakes, Karpathy inspired.
AI sucks. AI is great. I think that AI simply doesn't know its place. SYNTH tries to solve this, creating a "novel" system of classifying whether a project is vibe coded and _receives_ AI generated…
Buscador de imóveis com mapa, marcações e multi-portais