The second session in the series. A 60-minute working session on why most prompts get weak results, and a concrete structure for fixing them. By the end, attendees can recognise the six ways a prompt typically fails before sending it, apply a shared Prompt Standard skeleton, and tell a strong prompt from a vague one at a glance.
Audience: anyone who’s used Claude (or any AI tool) and felt the output was disappointing.
Topics covered
- What prompt engineering is — and what makes a prompt good.
- The ideal prompt format — a shared Prompt Standard skeleton, copy-paste ready.
- Why each block matters — what breaks if you skip a block.
- The six ways prompts fail — the failure patterns we kept seeing internally.
- Before / after — a real fix on a real document: a four-step chain that takes a Battery Warranty T&Cs document from a naive prompt to refined translations in French, Turkish, and German, with a reviewer agent in the loop.
- Save your good prompts — and how to ask Claude to turn one into a reusable skill.
- The
repromptskill — auto-check a prompt against the standard, and chain prompts for high-stakes work.
This session is also where the internal reprompt skill — prompt-quality scoring plus rewrite — got its public debut.