Prompting Mastery
For analysts, developers, and senior individual contributors who already use ChatGPT, Claude, or Copilot every day and want to move from ad-hoc prompts to repeatable, evaluable patterns. Eight chapters: principles that separate amateur from professional prompts; reliable JSON via structured outputs; chain-of-thought where it earns its keep; few-shot patterns that generalise; eval-driven iteration instead of vibe-checking; library structure, ownership, and versioning; the adversarial / prompt-injection chapter with real 2025 CVEs (EchoLeak, CamoLeak, Slack AI); and the close — the four things that decide whether the library lives past month three.
8
Chapters
~55 min
Duration
Intermediate
Level
No
Certification
Course Content
Prompting principles
The five things that separate amateur prompts from professional ones — and the chat-vs-reasoning model toggle that changes which principles dominate.
Structured outputs (JSON)
Stop parsing free text. Build a schema, force JSON Schema strict mode, and watch the model return parseable JSON on the first try every time.
Chain-of-thought reasoning
A 2×2 of when chain-of-thought earns its latency cost — and when adding "think step by step" makes outputs worse, not better.
Few-shot patterns
Examples that generalise vs. examples that memorise — a similarity scorer that shows the precise distance between your few-shots and your actual input distribution.
Evaluating outputs
A mini eval runner you can drop into CI tomorrow — happy path, edge cases, adversarial inputs — that turns "looks good to me" into a measured pass rate.
Prompt libraries and reuse
Four files per prompt — prompt, schema, evals, changelog — plus owner, versioning, and CODEOWNERS. The structure that survives turnover.
Edge cases and adversarial prompts
EchoLeak (CVE-2025-32711), CamoLeak, and Slack AI — the three 2025 incidents that prove prompt injection is now a production problem. The four defences that actually work.
Making your library stick
Owner, weekly review, evals in CI, frictionless contribution — the four things that decide whether the library lives past month three. The 4-week rollout plan, downloadable per team size.