1TommyCheung/hydra-swarm ? reverse-engineered prompt

Reverse engineered prompt

Build me a local multi agent dev harness that can plan tasks, spin up isolated git worktrees for each task, dispatch work to different coding agents, and only promote changes after they pass real checks.

I want the main lead agent to coordinate everything, keep the state deterministic, and treat the workers like untrusted processes. Each task should go through a clear trust boundary, first schema checks, then git evidence, then ownership checks, then sandboxed verification before it can be accepted. It should support Claude, Codex, OpenCode or GLM, and Kimi when they are available, but gracefully fall back if some are missing.

Make it work as a local tool with commands for starting a run, creating worktrees, dispatching tasks, checking status, promoting candidates, reviewing them, squashing accepted work, integrating in order, and cleaning up old worktrees after the run. Please include a doctor style preflight check, live progress visibility, and a clean audit log for each run. If you need to look up current docs online, do that.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab