AI & Agent WorkflowsSoftware EngineeringOpen accessPublished 3 Oct 2026
Delegate work to a different AI coding agent harness — codex, gemini, pi, opencode, crush, or aider — and verify what comes back. Two modes: `review` runs read-only in the live tree for a second opinion from a different model family; `build` runs write-enabled inside a throwaway git worktree so the result arrives as an inspectable diff on its own branch. Use when: delegate to another agent, second opinion from another model, ask Codex, ask Gemini, run this with opencode, cross-check with a different AI, offload work to save Claude quota, run several agents in parallel, compare how different models solve the same task, agent handoff, external agent, other CLI agent. Never …