7 Core Prompting Rules for Claude 3.5 Opus
JafarNajafov · x · 2026-07-23
For optimizing Claude 3.5 Opus (mistyped as Fable 5) in agentic workflows, the author summarizes 7 core prompting rules:
- Set effort explicitly (default: high)
- Act when you have enough info, don't survey options
- Do the simplest thing that works
- Audit every progress claim against a tool result
- Never explain your reasoning as text
- Log lessons to a memory file
- Lead with the outcome
More from coding & agent
- Code review must become an agent-to-agent pipeline, not a human-only loop — bibryam · 2026-07-23
- A meme contrasts happy vibe coding with miserable vibe debugging — miniapeur · 2026-07-23
- Cursor private workers can run on SSH VMs, screenshot shows — davidcrawshaw · 2026-07-23
- An agent hit GitHub Actions limits, spun up a VM, and kept the build running — davidcrawshaw · 2026-07-23
- Async agents fail when users think a 90-second job has frozen — Specific-Point-4026 · 2026-07-23
- Open-source skill scans your codebase first, then writes a README — Shruti_0810 · 2026-07-23