Courses

The Handbook

Eleven illustrated sections on who decides the next step at run time.

Prerequisite

None. All examples are scripted and work without a model key.

Outcome

Choose how much run-time control a problem should hand to a model.

Artifact

A completed 10-brief Control Room result and targeted sections to revisit.

Evidence

Completion means submitting all 10 briefs, not merely opening sections.

The Lab

Write a prompt, then score it against twenty real cases and watch the number move.

Prerequisite

A low-credit DeepSeek key for live stages; the rules exercise itself is local.

Outcome

Turn a brittle rule set into a prompt and evaluate it across varied cases.

Artifact

Rules, a prompt draft, preview evidence and one complete 20-case Eval result.

Evidence

A complete Eval is recorded at any score; 16/20 is a suggested target, not completion.

Build an Agent

Nine guided stages (0–8) on your own machine, plus the Stage 9 transfer project: the agent loop, a permission gate, prompt injection.

Starts with an on-site setup guide, then opens the local TypeScript course on GitHub.

Prerequisite

Basic TypeScript, Git, Node.js 24, a terminal and an editor.

Outcome

Build a bounded tool-use loop with Evals, approvals, review and trust boundaries.

Artifact

A Stage 9 Markdown design, Eval JSON, evidence and retrospective for your own domain.

Evidence

Local checker results plus the Stage 9 rubric; progress stays in course/progress.json.