Automate a second check on Claude Code replies

They built Sage to forward Claude Code’s answers to Codex so they could review another model’s take without manual copy-pasting.

Done with Claude CodeCodex· Advanced

What they did
They had been pasting Claude Code’s responses into Codex by hand to get a second opinion. To remove that step, they made Sage, which runs in a separate terminal, follows the main coding session as it happens, and sends each reply to another model for review. It keeps its own ongoing thread, and the user decides whether to use the feedback. Unlike an MCP setup they tested, it can also inspect the codebase, use tools, and search the web.
How it went
They said the setup cut down on bugs and surfaced bad plans earlier, but they did not share numbers. People in the comments said it was already catching useful issues, though the poster also mentioned some rough spots with plan mode and having to restart after Claude’s /clear.
Tips
If your agent can lose context, plan for a restart path and a way to reconnect to the new conversation. Keep the final decision with the human, especially before applying any suggested change, sending anything, or making repo edits.

Try it yourself with Claude Code

Build a small helper for [project path] that watches each response from [primary coding agent] and sends it, together with the current question and relevant code context, to [second model] for review. Show the two answers side by side, point out disagreements, and pause for my approval before you install anything, submit anything, delete anything, or change files. Make it runnable with one command on a sample question first, then let me choose whether to apply the review.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on reddit.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.