Codex CLI review: model seems good, but the CLI tooling has rough edges
A developer who mainly uses Claude Code and OpenCode tried after hearing praise following the GPT-5.5 release. The underlying model seemed capable, but the CLI harness itself had several practical problems. Its edit tool didn't work, forcing the model to fall back on bash just to modify files.
There's no way to expand a to see the full command and output, so it's impossible to inspect exactly what's being fed into context. There's no fullscreen mode. Unlike Claude Code, it lacks a , so the model had to use systemctl as a to start a server.
There's also no way to inspect what context a has or interact with it directly.
Key points
- Edit tool didn't work; model had to use bash as a
- No way to expand full command/output of to inspect context
- No fullscreen mode
- No — had to use systemctl to run a server
- No way to inspect or interact with context