I have evolved a similar 4-terminal approach which works excellently. Usually I have 2-3 projects going at the same time, as many claude code sessions, and often a custom llm tool i built in the terminal for rapid review using gpt5 with high effort and feed back to claude to iterate. Llm tool also generates docs and tests, which claude then debugs if necessary.
I find myself spending an hour in the morning just building context for the various sessions and tasks. I make a directory of markdown files with the specifics and references needed for the task, and feed them in one at a time. These are supplemented by the gpt-5 code reviews from the llm cli tool.
I've been trying to figure out the best interface to pull all of it together. The right HUD is really needed and I hope some crafty developer figures this out for agent management soon
I have evolved a similar 4-terminal approach which works excellently. Usually I have 2-3 projects going at the same time, as many claude code sessions, and often a custom llm tool i built in the terminal for rapid review using gpt5 with high effort and feed back to claude to iterate. Llm tool also generates docs and tests, which claude then debugs if necessary.
I find myself spending an hour in the morning just building context for the various sessions and tasks. I make a directory of markdown files with the specifics and references needed for the task, and feed them in one at a time. These are supplemented by the gpt-5 code reviews from the llm cli tool.
I've been trying to figure out the best interface to pull all of it together. The right HUD is really needed and I hope some crafty developer figures this out for agent management soon