# Risk of using Auto: Claude Code is reading config files outside the project folder, exposing API keys, and failing to detect the leaks when auditing. — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/f9ba35db2d528390

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-11-15 21:40 ingested 2026-11-16 02:38

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xx7ksq/risk_of_using_auto_claude_code_is_reading_config/)

## retained text



```
While delegating tasks to Codex, I caught Claude reading files outside my project folder (~/.codex/config.toml), which resulted in it dumping my Stitch MCP server's Google API key directly into the chat transcript.

Worse: when I asked another instance of Claude to scan all my project histories for similar boundary violations or leaked keys, it repeatedly told me none existed. It only admitted to reading the key multiple times over the past two weeks after I manually pointed it to the exact thread.

This leaves me in a catch-22: the agent won't find or admit to a secret leak unless I specifically point it out, meaning I have no reliable way to know what other keys it might have quietly compromised across my system.
```



document fetched 2026-11-16 00:35 · canonical https://reddit.com/r/ClaudeCode/comments/1xx7ksq/risk_of_using_auto_claude_code_is_reading_config
