An "anti Claude-speak" system prompt: active voice, no em dashes, no startup clichés
dressinbrass · reddit · 2026-09-10
A Reddit user shared a system prompt for suppressing Claude's characteristic writing style, reporting good results.
Rules: active voice with named actors, imperatives for rules, verbs over abstract nouns, one idea per sentence, concrete names/numbers over categories, no em dashes, cut filler words like "actually" and "simply", and deduplicate across the whole document.
Never list: no "not X, but Y" definitions, no rhetorical number seesaws, no tautologies, no fragment headlines or empty mottos like "X is the moat", no startup landing page jargon such as "wedge" or "ship cadence".
A rhetorical device is kept only when the plain version loses actual content.
More from coding & agent
- Full pipeline: niji + GPT Image + local MiniMax H3 + Codex for pixel-perfect animation — Hailuo_AI · 2026-09-10
- Instinct launches agent-to-agent protocol so personal AIs can coordinate your plans — mon__lim · 2026-09-10
- Traser: a local deterministic tool to triage agent runs that succeed but return wrong results — Sensitive-Parsnip-12 · 2026-09-10
- Experiment: with a false premise and stale artifacts, only one of three agents found its peers — turtle_bazon · 2026-09-10
- Iron Man fan builds interactive suit teardown site with GPT-6 Astra, Three.js — TheMoonMidas · 2026-09-10
- Hermes Agent routing flaw silently switched user to metered OpenRouter, costing $100 — WolframRvnwlf · 2026-09-10