Leo Zhang builds AI tools, backend systems and interfaces for organizing work. His Oh My Subagents project makes delegation between coding agents durable: responsibilities, waits and accepted results survive beyond a single conversation.[1][2]
Start with I let an subagent workflow refactor my codebase for three days →
GitHub 
X Background
Zhang’s portfolio describes work spanning system design, implementation, testing and deployment. He is completing a Master of Computer Science at the University of Sydney. His projects include a browser tab organizer and ImageSage, a Windows image gallery with semantic search and file-management features.[1]
Contributions
Oh My Subagents records assignments, task ownership, waits and checkpoints in a local runtime. A manager can delegate work, wait for the child assignments and continue with their complete returns after an interruption. Provider completion is evidence for the parent; the task lead still decides whether to accept the result.[2]
Ideas
Its workflow definitions describe who is responsible, while managers choose the sequence of work from current evidence. The visual console and conversational operator use the same controller operations. This keeps the coordination record consistent between chat and the interface.[2]
Tools, writing & talks
Writing, research & talks
Start here
A local runtime for persistent coding-agent teams, assignments, waits and interruption recovery.
Requires Python 3.12+ and a configured agent provider. Read the bug-fix example and accountability contract; follow how child checkpoints become evidence for the parent, and distinguish the accepted result from a provider merely finishing.
Developer and maintainer[1][2]Original source