# How often has a coding agent done something you had to undo? — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/731e014823cd694a

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-10-31 05:42 ingested 2026-11-01 02:37

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1xk277d/how_often_has_a_coding_agent_done_something_you/)

## retained text



```
For people running Claude Code and similar coding agents with auto-approve, in CI, or in the background (maybe even headless agents): what's the worst thing an agent did that you had to clean up after? How did you notice, and how did you fix it?



Also curious what you do to stay safe today: committing before every run? containers? --dangerously-skip-permissions and hoping, something else?



Was researching this space, hope to get some guidance from you guys!
```



document fetched 2026-11-01 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xk277d/how_often_has_a_coding_agent_done_something_you
