I built an open-source CLI called codex-workspace-bootstrap for a problem I kept running into with AI coding tools:
the repository says one thing, while the agent instructions say another.
For example:
- the repo uses pnpm
- .cursor/rules tells the agent to use npm
- AGENTS.md references a script that does not exist
- different agent config files use conflicting test or build commands
These issues are easy to miss until an agent has already started changing the codebase.
codex-workspace-bootstrap adds a deterministic preflight step before that happens:
cwb preflight .
It can detect:
- package-manager mismatches
- missing package.json scripts
- instruction drift across Cursor, Codex, Copilot, Claude, Gemini, Continue, and Cline
- nested AGENTS.md
- nested .cursor/rules/*.mdc
- risky tracked filenames
It is local-first and does not execute commands extracted from instruction files.
It also works as a GitHub Action, so the same checks can run in CI before changes are merged.
GitHub:
I’m especially interested in feedback from people using Cursor together with other coding agents in the same repository.
Things I’d like to hear about:
- false positives
- monorepo edge cases
- unusual Cursor rule setups
- instruction conflicts you’ve actually hit in real projects
The project is free and open source.