Claude Code VS Code prompts hide symlink destinations
Original title: Claude Code Extension Could Let AI Agents Access Files Outside Trusted Workspaces Through Symlinks
Plugin 2.1.322 remains unpatched after Anthropic closed the report; the CLI now shows real paths.
A researcher says Anthropic’s Claude Code extension for Visual Studio Code shows only a workspace-relative path when an agent follows a symlink outside the project. Users can approve reads or writes, including to locations such as SSH keys, without seeing the real destination. Anthropic closed the report as informative. The researcher says plugin version 2.1.322 is unpatched, while the Claude Code CLI was later changed to resolve symlinks before approval.
Unconfirmed: Anthropic says this is not a vulnerability and is outside its permission threat model; the researcher disagrees.
- The extension has over 30 million Visual Studio Marketplace installs.
- Adversa confirmed related symlink behavior in several other agent CLIs.
- Git-tracked symlinks could reach colleagues’ workspaces after a sync.
- Before approving, run realpath on the displayed path and deny unexpected destinations.
- Adversa advises failing pull requests that change agent or MCP configuration access.
Impact: Agent file reads and writes outside the trusted workspace
| Product | Affected |
|---|---|
| Anthropic Claude Code VS Code extension | plugin version 2.1.322 |
| Anthropic Claude Code CLI | — |
- 1.Run realpath on the displayed path and deny approval if it resolves outside allowed locations
- 2.Scan repositories for tracked symlinks that target configuration files
- 3.Isolate CI runners handling untrusted code from runners holding production secrets
- Vendors
- Anthropic
- Products
- Claude CodeVisual Studio Code
- Organizations
- AdversaUltrastrikeHackerOne
- AI models
- Claude
From the original article
9 min read at gbhackers.com
Anthropic’s Claude Code extension for Visual Studio Code reportedly shows misleading file paths in its permission prompts. A researcher says this can lead users to approve file access outside their trusted workspace. Anthropic closed the report as informative and did not treat it as a vulnerability.
By Mayura Kathir · gbhackers.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.