Claude Prompt Engineering Docs

What it does

• Routes before it teaches: a 'when to prompt engineer' check (model choice sometimes beats prompt tweaks) and a 'before prompt engineering' prerequisite list — success criteria, empirical tests, a first draft
• Model-specific prompting guides for the current lineup: Fable 5.1, Fable 5, Opus 5.5, Opus 5, Opus 4.8, Sonnet 5.5, Sonnet 5 and Haiku 5.5
• Prompting best practices is the living techniques reference — clarity, examples, XML structuring, role prompting, thinking, prompt chaining
• Chains into the surrounding evals flow (define success criteria, build evaluations) and a metaprompt-based prompt generator notebook from the Claude Cookbook

When to use it

The canonical starting point when a prompt underperforms: use the overview to pick the right layer — technique page, model-specific guide, or the evals flow — instead of reinventing tricks from blog posts.

Getting started

No install — it is a documentation site. Start at the overview, then keep Prompting best practices open as the living reference; don't skip the 'Before prompt engineering' checklist (success criteria, empirical tests, a first draft).

Things to know

The overview itself is a hub, not a handbook — the substance lives in the child pages (best practices, per-model guides, evals). Techniques are tuned per model: a trick that helps one Claude generation can be neutral or harmful on another, so match the page to the model you actually call.