I have 3 go-to dinner spots in NYC: Nyonya, Izakaya Zest, and Joe’s Shanghai. My girlfriend vetoed all 3 one night, which was confusing because historically these picks have a pretty good success rate. Then she mentioned one small detail: we were walking around Little Island after dinner. Oh. LOS TACOS No. 1. The recommendations weren’t bad. I was just working with incomplete context. That’s basically the idea that led me to PlaybookDiff. I kept noticing that Claude Code and Codex could be sitting in the same repo, but because they discover instructions, skills, and MCP config differently, they weren’t necessarily working from the same playbook. So I built a small open-source tool to make those differences visible. You can run it locally, use it in CI with the GitHub Action, or just poke around the repo. Would genuinely love feedback from anyone using both Claude Code and Codex, especially if you’ve run into weird config drift or “why did this agent do that?” moments. playbookdiff.dev

PlaybookDiff

Keep Claude Code and Codex on the same playbook

Keep Claude Code and Codex on the same playbook