Agent-Readable Stubs Make Draft Courses Operable

·https://wizardshit.ai/the-hive-multi-agent-orchestration
articleaiagent-loopsorchestrationcmuxpi-cmuxcoursesagent-discovery

maps to joelclaw content publishing: ship markdown twins, manifests, and companion skills before the full course exists.

The Hive: Multi-Agent Orchestration is still a draft course on wizardshit.ai, but the interesting part is that the route is already useful to agents. It has a real page, a markdown twin, course metadata, a 10-lesson outline, and a companion skill install path from badass-courses/wizard-skills.

That’s the clever bit: the course isn’t waiting for polish before it becomes infrastructure. The stub says exactly what the thing is supposed to become: a course about using cmux as a control plane for visible, steerable, auditable fleets of AI agents, with pi-cmux acting as the bridge between terminal surfaces and agent lifecycle tools.

This is useful for joelclaw because it treats educational content like an agent-operable system, not a pile of pages. A draft can expose routes, manifests, markdown, companion skills, and teaching instructions first. Then humans and agents can both build against the shape while the actual lesson text catches up.

It also leaves a nice breadcrumb for agent discovery: the HTML page advertises the markdown version through an agent hint, and the markdown version tells a teaching assistant how to fetch lesson content with Accept: text/markdown. That’s not content marketing fluff. That’s a working interface.

Key Ideas

  • A draft course can still have a real public route, markdown projection, metadata, and companion surface before the lessons are finished.
  • cmux is framed as an agent control plane: workspaces, panes, monitoring, steering, recovery, and fleet visibility from the terminal.
  • pi-cmux is the bridge layer that makes the terminal multiplexer agent-aware with sidebar status, lifecycle hooks, and tool activity.
  • The lesson outline names practical failure modes: agent fork bombs, shell escaping, designs collapsing in contact with reality, completion detection, and attention routing.
  • The page’s markdown twin gives agents a cleaner ingestion path than scraping rendered HTML, which is exactly the kind of boring-but-important interface agent-discovery needs.

Links

← All discoveries