Forced rm detection tightened in Codex
OpenAI's Codex team said it has rolled out changes over the past couple of weeks that cut the risk of Codex running destructive commands while it works. The effort started with an investigation into a small number of reports.
Per the team's post on GitHub, forced rm detection got stronger. It picks up literal forced rm invocations in control flow, substitutions, pipelines, traps, nested shells, and behind sudo or env wrappers.
Up next
Every Codex turn fails on gpt-5.6 Sol after 0.148.0
Per a bug report, every turn in Codex CLI 0.148.0 fails on gpt-5.6-sol, because the client keeps sending prompt_cache_retention and the model rejects it. Turns that already started tool calls hang instead of erroring.
Related stories
- Prompt rule sends Codex to the web for a local file
- GitHub report: Codex memory sends chats cross-provider
- A dependency wrote AGENTS.md and Codex hid the edit
- Codex CLI 0.157.0 leaves a daemon running after you quit
- Codex Security Cloud reviews commits with the laptop closed
- Reading a repo was enough to break out of Codex
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.