Skip to content
Cyber Press·Indexed by this site · Originally published· Event →· Its place in the scan record →

Researchers say Claude Code VS Code extension symbolic link prompt may hide paths outside the workspace

Claude Code VS Code Permission Prompts Can Conceal Access to Files Outside the Workspace

AI summary

Security researchers found that Anthropic's Claude Code VS Code extension displays misleading permission prompts: the path appears to be inside the current workspace but actually points outside it via a symbolic link, and after the user approves, the agent may be able to access sensitive files elsewhere on the system. The researcher said the issue remains unfixed in version 2.1.322 of the plugin; Anthropic closed the report as informational, saying the prompt itself is sufficient, the information shown in the prompt is not a security control, and symbolic links are not within Claude Code's permission threat model. The finding follows Adversa AI's May 2026 CLI research: according to Adversa, the CLI at the time displayed only the symbolic link path, and Anthropic rejected that report but quietly hardened it weeks later so that the prompt shows the resolved real path.

Read originalcyberpress.org