Agent-orchestrating multiplexer
The inversion of the agent-native-multiplexer: an agent drives the multiplexer. Via herdr's official agent-skill-file, it can create workspaces/sessions, start other agents inside them, prompt them, and read their replies. alejandro-ao's demo: Tau receives "create a new Herdr session, run Pi inside of it and ask it to say hello world" and executes the full chain. A DIY sub-agent runtime built from a terminal tool โ the hobbyist edge of what dhh does with a personal bot system and what harnesses do natively.
Cross-harness version (dave-ebbelaar, Sept 2026): with the skill installed globally for Grok, Codex and Claude, a Codex agent at medium effort is told to "kick off another agent using the herder skill" in a new tab with Opus 5 to review the codebase. It reads the skill file, opens the tab, starts Claude Code and types the prompt. The tabs can message each other through the CLI, so a finished agent reports back and the orchestrator resumes.