Orchestrate a multi-agent swarm to build verified, falsifiable "concept bundles" — teaching units where every claim derives from ONE runnable ground-truth file whose output is pasted verbatim into a guide. Use when the user wants to build/extend a learning curriculum (language concepts, algorithms, system design, interview patterns), spin up subagent workers to produce runnable + markdown (+ optional interactive HTML) lessons, run a verification sweep, or scaffold a new learning repo section. Handles BOTH interactive (.html companion) and non-interactive (runnable-only) flavors under the same orchestrator+worker discipline.
2026-06-28