NewApp Monitoring is liveSign up free
AgentbeamBeta
AI agent vulnerability · Claude Code

GHSA-ph6w-f82w-28w6: Claude Code Vulnerable to Arbitrary Code Execution Due to Insufficient Startup Warning

highCVSS 4.0 8.7

Before Claude Code 1.0.87, the startup prompt "Do you trust the files in this folder?" did not say that choosing "Yes, proceed" lets Claude Code execute files in that folder without further confirmation. A user could accept it on an untrusted repository without knowing the risk. Version 1.0.87 clarifies the warning. Auto-updating installs already have it.

Published Sep 3, 2025 · Analysis Nov 1, 2026

What an attacker can do

This is a consent flaw, not a parsing bug. When Claude Code starts in a new directory it asks whether to trust the folder. The old wording hid what "Yes, proceed" grants: permission to run files from that folder with no later prompt. A repository can ship project configuration and scripts. If a developer opens a cloned, untrusted repository and accepts the prompt, those files can run code on their machine.

The attacker needs no access to the victim. They need the victim to open a hostile folder, for example a repository from an unknown source, and click through the trust dialog (the vector marks passive user interaction). The result is code execution as the developer, with their files, tokens and cloud credentials. The advisory is thin on mechanism. It names no specific file or setting, and says only that the fix was to the warning text.

Who is affected

@anthropic-ai/claude-code before 1.0.87. Check the installed version:

claude --version
npm ls -g @anthropic-ai/claude-code

Users on the standard auto-update channel received the fix automatically. Pinned installs, containers and manually updated machines may still run an older build.

@anthropic-ai/claude-code

Ecosystem
npm
Affected
< 1.0.87
Fixed in
1.0.87
Check
npm ls @anthropic-ai/claude-code

How to fix it

Update to 1.0.87 or later:

npm install -g @anthropic-ai/claude-code@latest

The fix changes the trust prompt so it states that "Yes, proceed" allows Claude Code to execute files in the folder without additional confirmation. The advisory lists no other workaround. Treat the clarified warning seriously: answer yes only for folders whose contents you have reviewed or trust, and not for freshly cloned third-party repositories.

How to tell if you were hit

The advisory gives no indicators. Check which machines run a Claude Code build older than 1.0.87 with claude --version. On a suspected machine, look at repositories opened with Claude Code from untrusted sources, and at shell history and process lists for commands launched from them. Beam records agent shell commands and file access, so you can review what a Claude Code session ran after a folder was trusted.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

When Claude Code was started in a new directory, it displayed a warning asking, "Do you trust the files in this folder?". This warning did not properly document that selecting "Yes, proceed" would allow Claude Code to execute files in the folder without additional confirmation. This may not have been clear to a user so we have updated the warning to clarify this functionality.

Users on standard Claude Code auto-update will have received this fix automatically. Users performing manual updates are advised to update to the latest version.

Thank you to https://hackerone.com/avivdon for reporting this issue!

Questions

Does this have a CVE?

No. The advisory GHSA-ph6w-f82w-28w6 lists no CVE id. It is rated high with CVSS 4.0 score 8.7 and CWE-94.

What changed in 1.0.87?

The trust prompt wording. It now explains that selecting "Yes, proceed" lets Claude Code execute files in the folder without further confirmation. The advisory does not describe other code changes.

Do I need to update manually?

Not if you use standard auto-update, which delivered the fix. If you update by hand or pin a version, move to 1.0.87 or later.

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

Coverage on the Beam blog

More Claude Code vulnerabilities

All Claude Code entries