A harness gives the model a working environment: tools, context, permissions, and a repository. I still choose the outcome and assess the result.

Solid arrows show the T3 fork lineage. The dashed connection shows Kitshn integration. Yarden Zamir’s Kitshn is a separate project, alongside his work on T3.

The thread connects the conversation to the project and its changes. A separate worktree is available when a task needs an isolated checkout.

This is an example workflow, not a requirement to parallelize every task. Research can inform implementation; review examines the resulting changes; the main agent resolves the findings.

These are example machine roles, not an inventory of my hardware. The linked environment has its own checkout and tools; delegation respects the grant and calling-thread limits.

Selected workflow failures can route into a repair task in my fork. A repair proposal still needs validation and the applicable review and merge process; this is not a guarantee of automatic recovery.

The change is in the handoffs: less context to reconstruct, fewer manual status checks, and a clearer point for my decisions. The lanes are conceptual, not durations or a productivity benchmark.