Tools and subagents I use: plan mode.
Most harnesses already ship this out of the box, and there are great advantages to using this tool. The biggest one is freeing up the context window between the planning and execution phases.
The idea is that you can spend a lot of your context window on research until you agree on an implementation, then reset the context for that implementation. It’s well known that agents perform better when their context is clear, so having this routine helps.