▲ Claude Academy
Sign in

Vault note

Track 2 — Effective Prompting

Track 1 taught you to talk to Claude; this track teaches you to get reliable, repeatable results — the difference between "it worked that time" and "it works every time." You'll learn structured prompting techniques, apply them to messy real-world tasks, and then learn to measure prompt quality with basic evaluations so you're improving on evidence, not vibes.

Audience: anyone who wants dependable outputs (no coding required). · Est. time: ~6–8 hours over 4–5 sessions. · Prereq: track-1-claude-foundations. · Outcome: reliable prompts plus a basic evaluation workflow.

Setup: nothing beyond a claude.ai account (setup-checklist, section b). The classic prompt tutorial also runs free in your browser.

Sequence

1. Prompt engineering, structured · ⏱ ~2.5–3 h

  • Course: prompt-engineering-interactive-tutorial — the canonical techniques: clear and direct instructions, examples (few-shot), assigning a role, using XML tags to structure input/output, giving Claude room to think (step-by-step / chain-of-thought), prefilling, and controlling format.
  • Project: p02-prompt-lab — build a small personal "prompt lab": take three weak prompts and systematically rewrite each using these techniques, recording what changed and why.

2. Real-world prompting · ⏱ ~2 h

  • Course: real-world-prompting — applying the techniques to genuinely messy tasks: long documents, ambiguous requirements, multi-step jobs, and prompts you have to maintain over time. Emphasis on iteration and on writing prompts that survive contact with real inputs.

3. Measure it — prompt evaluations · ⏱ ~2 h + project

  • Course: prompt-evaluations — why eyeballing doesn't scale; building a small test set, choosing metrics, grading (exact-match, code-graded, and LLM-as-judge), and iterating prompts against scores instead of impressions.
  • Project: p09-eval-harness — build a minimal eval harness: a handful of test cases, a grading method, and a before/after score showing a prompt you measurably improved.

You're ready to move on when…

  • You can take a vague prompt and improve it with at least four named techniques (role, examples, XML structure, step-by-step, prefill, format control).
  • You can structure inputs and outputs with XML tags so results are parseable and consistent.
  • You can explain when to use few-shot examples vs. a clearer instruction.
  • You can build a small eval set and judge two prompt versions by score, not gut feel.
  • You've completed p02 and p09.

See also