Claude Code VS Code extension hides symlink targets in prompts
Original title: Researchers Reveal How Claude Code VS Code Extension Can Access Files Outside Trusted Workspaces
Version 2.1.322 still omits the real path; Anthropic closed the report and no patch is available.
Ultrastrike researcher Carl Pearson says the Claude Code extension for Visual Studio Code can ask permission using a workspace path while a symlink points the file outside that workspace. A user may approve what looks like a safe write, including toward locations such as a home folder that could expose SSH keys. Anthropic reviewed the HackerOne report, closed it as informative, and does not treat the behavior as a security issue. The command-line tool now shows real locations, but the extension had not received that change as of version 2.1.322.
Unconfirmed: Anthropic says the prompt text is not a security control and that symlinks fall outside its permission threat model.
- The extension has more than 29 million installs.
- Ultrastrike published the finding on November 11, 2026.
- An earlier CLI display flaw was reportedly fixed quietly after May 2026 research.
- Pearson says the extension can rename files without a prompt in some modes.
- Ultrastrike advises running realpath on the displayed path before approval.
Impact: Approved file access may reach paths outside the trusted workspace
| Product | Affected |
|---|---|
| Anthropic Claude Code VS Code extension | 2.1.322 |
- 1.Before approving, run realpath on the path shown in the permission dialog.
- 2.Deny the request if the resolved location is outside the workspace.
- 3.Review repositories for unexpected symlinks after each sync.
- Vendors
- Anthropic
- Products
- Claude CodeVisual Studio Code
- Organizations
- UltrastrikeAdversa AI
From the original article
4 min read at cybersecuritynews.com
Security researchers have found that the Claude Code extension for Visual Studio Code can show a permission prompt that hides where a file really sits. A user may approve access to a path that looks safe, while the file lives outside the trusted workspace. Anthropic reviewed the report and does not treat it as a […]
By Tushar Subhra Dutta · cybersecuritynews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.