Done. Here's the verdict:
This repo looks like a legitimate Anthropic Claude Code source distribution. Nothing malicious detected. The 180 extra files vs your other copy are almost certainly the bundled plugins โ 13 plugin directories in plugins/ that ship with the source but wouldn't be in a minimal/npm install.
Summary of what I checked:
- โ No obfuscated code โ no hex-encoded payloads, no unicode tricks (Cursor.ts false positive is just Intl segmenter imports)
- โ
No secrets/credentials โ no
.env, no API keys, no private keys - โ
No postinstall hooks โ
package.jsonhas no lifecycle scripts - โ No exfiltration โ network calls are all standard (Anthropic API, Sentry, Statsig, GitHub)
- โ
No
.gitdirectory โ this is an extracted archive, not a cloned repo
Things worth noting (not malicious, but interesting):
-
ralph-wiggumplugin โ a Stop hook that intercepts Claude's exit and feeds its output back as input in an infinite loop. By design from Anthropic (Daisy Hollman). Has a completion promise mechanism to eventually exit. -
security-guidancehook โ writes debug logs to/tmp/security-warnings-log.txt. Benign but worth knowing if you ever run it.