openai/symphony ? reverse-engineered prompt

Build me Symphony. It turns regular project tasks into fully isolated autonomous runs so my team can focus on managing the actual work instead of hovering over coding agents the whole time.

Have it watch a task board like Linear for incoming items then automatically spawn an agent for each one. The agent handles the implementation end to end and returns clear proof of work such as passing CI checks, review notes, complexity scores and a short walkthrough video. Once everything looks good the agent submits the PR and lands it safely without us needing to babysit every step.

Follow the core idea from the openai symphony project and make it work well for codebases that already use harness engineering. Use whatever language and setup feels clean and reliable to you.