does openhands currently support a teams setup like the current version of claude? I'm finding that paired with subagents to work better than anything I've ever seen. I have been able to go through processes I do every day, take a day or so to create a high quality team with tests and feedback loops and fully automate them. The quality has been higher than when I code things manually to be honest. I have not figured out how to duplicate the claude code team setup in anything outside of claude code though. I've made it as token efficient as I can think of and use sonnet when possible, but I can still burn through my alotment on a max plan in about 20 hours of work. it's essentially these roles with hooks to make sure they can't continue without the adversary and tester checking things. Role Responsibilities | Role | Does | Does NOT | |------|------|----------| | Auditor | Surveys codebase, forms hypothesis, writes proposal, negotiates with adversary, hands off to implementer | Implement code, run tests, make verdicts | | Implementer | Receives approved proposal, implements changes, coordinates tester, makes verdict from numbers, writes journal | Propose ideas, negotiate with adversary, interpret results creatively | | Adversary | Reviews proposals for flaws, checks overlap between cycles, rates concerns BLOCKING/IMPORTANT/MINOR | Implement fixes, run tests, override test results | | Tester | Runs smoke tests, adversary-requested tests | Interpret results, recommend actions, modify test configs | | Team Lead | Orchestrates spawning, stagger timing, merge ordering, shutdown, budget tracking, journal assembly | Propose changes, review code, run tests | the issue is if I just use subagents for this and don't have a manager then they start drifting and doing conflicting things. the manager keeps them on task. the whole teams setup with the experimental flag in claude code was the big unlock. an automated workflow was too unreliable or would run out of context before that.

Yes somewhat, and we’re working on bringing them even closer to parity! See our Claude parity plans here, we’d love any comments on the specific issues that you think are most important: https://github.com/OpenHands/software-agent-sdk/issues/2055