System prompts beat user prompts: taming verbose Claude Opus 5
IndyDevDan · youtube · 2026-08-17
IndyDevDan argues Claude Opus 5 is one of the best models ever but one of the worst to talk to — verbose, stuffed with filler, torching output tokens. His fix isn't a new model but prompt engineering, focused on the system prompt rather than user prompts.
Key techniques:
- Positive + negative patterns: state exactly what to replicate and what to avoid (no "load bearing", em-dash chains)
- Reference points: D1/R1/F1-style codes build a shared shorthand and cut repeated tokens
- Hard operational boundaries: deliver only what was requested — no scope creep or surprise refactors
- Aliases: SCR/ELI/FOC micro-skills that expand into full instructions on demand
- In-context distillation: real do/don't examples to lock in communication patterns
The system prompt applies across every exchange, making it the highest-leverage place to control verbosity, cost and quality regardless of which SOTA model ships next. The fix repo is open on GitHub.
More from coding & agent
- Meta open-sources Muse Spark Open agent model with day-0 TPU support in SGLang — BanghuaZ · 2026-10-16
- Saying "load-bearing" in your prompt makes Opus 5.5 stop holding back — zacharynado · 2026-10-16
- You.com partners with Paperclip to give every agent live web access — PolarBearby · 2026-10-16
- I priced every Grok Bot call: $20 to $200 plan in 5 days, long jobs eat 52% — jonathanmalkin · 2026-10-16
- GitHub is criminally underrated as a free data store, devs argue — blelbach · 2026-10-16
- Claude Desktop doesn't support MCP sampling — here's a workaround for two-way MCP apps — wuj · 2026-10-16