crossagent pairs the coding agent in one tmux pane with the one next to it. One writes, the other pushes back, before you ever see the diff.
A coding agent reviewing its own work has the same blind spots twice. If you already run two harnesses side by side in tmux, a second opinion is sitting one pane over. You just had to paste between them by hand.
crossagent turns that pane into a reviewer. The driver asks over MCP, the peer reads the repo and answers with a verdict, and nothing ships without the other one having a say.
Both panes load the same server. Each one declares its pane, harness and repo.
The driver briefs the peer once: same-level copilot, reads the repo, never writes.
Every plan and diff goes through discuss. The reply ends in one verdict.
Every reply from the peer ends in one line: the word VERDICT and one of four calls. The driver reads that line before it does anything else.
The peer is a same-level copilot, not an approver. It reads the repo, runs read-only checks and disagrees when it disagrees. Looping to consensus and escalating stay with the caller. crossagent never decides for you.
VERDICT line.The driver asks again.claude mcp add crossagent -- npx -y crossagent-mcp codex mcp add crossagent -- npx -y crossagent-mcp npx -y crossagent-mcp setup --link Using another coding agent? crossagent only needs a harness that runs in a tmux pane. Add yours and open a pull request.