Workflow Orchestrator

This is the orchestration layer that chains skills together automatically. When a skill hits a milestone like completing a PRD or finishing an implementation, this reads the hook configuration and fires off whatever comes next, whether that's logging the session, kicking off a code review, or triggering the self-improving agent to learn from what just happened. It supports three execution modes: auto for immediate fire-and-forget stuff like logging, background for async work like reflection, and ask_first for anything that needs confirmation like creating PRs. The universal learning pattern is clever, every skill completion can trigger self-improvement in the background so the system learns from its own executions. Saves you from manually running the next logical step in multi-stage workflows.