CVE-2026-55607

A vulnerability in Claude Code (versions 0.2.54 to 2.1.163) auto-approves WebFetch requests to huggingface.co without a permission prompt or --allowedTools restrictions. An attacker can exploit this to send requests to attacker-controlled repositories, creating a covert channel for data exfiltration.

Claude Code before 1.0.93 has a parsing error in shell commands related to $IFS and short CLI flags, allowing bypass of read-only validation and arbitrary code execution. This requires adding untrusted content to the Claude Code context.

Claude Code, an agentic coding tool, prior to version 2.0.31 has an error in sed command parsing that allows bypassing the read-only validation and writing to arbitrary files on the host system.

Claude Code prior to version 1.0.39, when running on a machine with Yarn 3.0 or above, could be tricked to execute code contained in a project via yarn plugins before the user accepted the startup trust dialog. Exploitation requires starting in an untrusted directory and using Yarn 3.0+.

Claude Code versions 2.1.59 through 2.1.128 write responses from the /copy command to /tmp/claude/response.md without UID isolation, randomness, or symlink protection. The file is created world-readable (0644) in a world-traversable directory (0755), allowing any local user to read a privileged user's Claude response, which may contain secrets or credentials.

In versions 2.1.63 through 2.1.83 of Claude Code, the folder trust determination logic used the git worktree commondir file without validating its contents. An attacker could craft a malicious repository with a commondir file pointing to a path the victim had previously trusted, causing Claude Code to bypass its trust confirmation dialog and immediately execute hooks defined in `.claude/settings.json`. Exploitation requires the victim to clone the malicious repository and run Claude Code within it, and the attacker must know or guess a path the victim had already trusted. This issue has been fixed in version 2.1.84.