Developers: Claude Opus Needs Clean Slate and Goal-Oriented Use

Developer Matt Shumer argues that Claude Opus's actual capability far exceeds current public perception, but only if users change their traditional usage habits. This view has been echoed and endorsed by several other developers.

Confirmed

• Clear historical presets: It is recommended to delete all existing skills, MCPs, claude.md, and other configuration files, letting the model run from a clean state.
• Shift prompting paradigm: Stop hand-holding the model with specific execution steps. The recommended approach is to directly provide clear goals and background context, letting the model find its own path.
• Fine-grained control: If fine-grained control is needed, do not hard-code instructions; instead, collaborate with the model to produce a detailed plan document first, then let it execute.

Why it matters

• This finding reveals a potential "over-engineering" problem in current AI usage. Excessive preset scaffolding and micro-instructions actually limit the autonomous reasoning and planning capabilities of advanced models like Claude Opus. Shifting to goal-oriented interaction marks a paradigm shift in AI prompt engineering.

2026-08-08 ~ 2026-08-09 · 5 related posts

• [source] Matt Shumer's Tips for Claude Opus 5: Clear Presets and Let Go of Control — mattshumer_ · 2026-08-08
• [source] Using Claude Opus: Clear Presets and Give Goals for Better Results — trq212 · 2026-08-08
• [source] Prompting Paradigm Shift: Stop Prescribing Steps, Let Models Navigate — mattshumer_ · 2026-08-08
• Unlocking Claude Opus: Clear Presets and Only State Goals — MartinGTobias · 2026-08-08
• Stop Micromanaging Opus 5: Anthropic Engineer Advocates Goal-Oriented Prompting — daniel_mac8 · 2026-08-09