Symphony flips the usual agent workflow: instead of babysitting Claude while it codes, you define tasks in Linear and let isolated agent runs complete them autonomously. Each task gets its own branch, the agent implements it, opens a PR, and collects proof of work (CI status, reviews, complexity analysis) before you ever look at it. You review outcomes, not sessions. The skill includes an Elixir reference implementation that polls Linear, spawns Codex agents, and manages the whole PR lifecycle. It assumes you're organized enough to spec work clearly and trust agents to execute without supervision, which is a big leap but probably where this is all headed anyway.
npx skills add https://github.com/aradotso/trending-skills --skill openai-symphony-autonomous-agents