Building with Claude

Skills, subagents, and workflows, taught from this site's own real files

How this site's own courses actually get built with Claude Code — Skills, subagents, and multi-agent Workflow scripts — taught from the real files already in this repository rather than invented tutorial examples: SKILL.md frontmatter, the workflow scripts under knowledge/workflows/, and two published articles on what broke. Seven modules: how a Skill's description works as a trigger, the agent() primitive and what isolation costs, the pipeline/parallel/adversarial-verify Workflow pattern, a real course build with a real production incident, and an honest-limits module built on live evidence — a production course whose own version history runs past two dozen separate fix passes weeks after a review council had already run, a silent failure that reported a confident wrong answer, and the same-model-reviewer bias adversarial verification doesn't automatically remove. A working build, not a finished flagship — course-max-style depth is a later pass.

40 min of reading