Spin up a second agent as an independent QA reviewer
Codex found the user's Claude login, installed Claude Code in a sandbox, and used it as a QA agent without being told how, leaving credentials untouched.
Done withCodexClaude Code
- What they did
- While reviewing defects in the user's AI system, Codex searched the machine for Claude, found an existing Claude configuration and OAuth login, downloaded Claude Code into a temporary folder, fixed a Node runtime install issue, and launched Claude non-interactively as an auditor. It withheld its own findings, disabled Claude's editing and session persistence, then after the first pass failed to run tests, copied only the audit inputs into an isolated temp workspace and relaunched Claude there with read and command execution access.
- How it went
- According to the post, Claude's first review largely matched the defects Codex had already found, and the second sandboxed run could execute Python tests and adversarial probes. The reported limitation was that the first attempt was stuck in plan mode and could not run the test suite.
- Worth knowing
- The catch was environmental: this only worked because an authenticated Claude OAuth session already existed on the Windows user account, and Codex reused it without changing those credentials.
Try it yourself with Codex
After you finish implementing [feature or fix] in [project folder], set up a second, independent AI agent in a sandbox to act as a QA reviewer and test your work. Have it check [what to verify, e.g. edge cases, failing tests, UI behavior] and report problems back to you. Do not read, copy, or change any of my credentials, and finish with a summary of the QA findings and fixes.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · Aug 1, 2026
Related use cases
Turn tickets into merged pull requests
Pulls tasks from GitHub Issues or Linear, then runs Claude Code or Codex in isolated Kubernetes pods to carry each one through to a PR.
Design and build a robotic arm with no CAD experience
A builder used Claude Code with FreeCAD and Claude Vision to design, print, and assemble a 5-DOF servo arm with vision and voice control.
Run an entire business from the terminal with Claude Code
Poster built a system that runs their whole business through Claude Code instead of Claude.ai; the specific tasks aren't detailed in the snippet.
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Vibe-code and deploy an idle clicker game
Gilbert Pellegrom had his OpenClaw agent build a simple idle game from a conversation, commit it to GitHub and auto-deploy it to Vercel.
Ship more pull requests with Devin across tools
Ryan Carson uses Devin from iPhone, Slack, browser, GitHub and Linear for real work and reports higher PR shipping velocity.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.