The vulnerability is a path traversal flaw in the OpenClaw sandbox skill mirroring feature, where the name field supplied in a skill frontmatter is not sanitized before being written to disk. An attacker can supply a crafted skill package with traversal sequences such as “../” or absolute paths, causing the system to create or overwrite files outside the intended sandbox workspace. This can lead to arbitrary file creation or modification, potentially enabling the execution of malicious code or the compromise of system configuration files, depending on the permissions of the running process.

Affected systems include all OpenClaw installations running a version earlier than 2026.2.14. The vulnerability is tied to the OpenClaw product and is listed as affecting all builds before that release date, with no specific sub‑version constraints beyond the major release cutoff.

The CVSS score of 5.6 indicates a moderate impact, and the EPSS score of less than 1% reflects a low likelihood of widespread exploitation at present. The flaw is not listed in the CISA KEV catalog. Attackers would need to send a malicious skill package to a configured sandbox, implying a remote attack vector that requires the ability to upload or otherwise submit a skill. If the sandbox operates with elevated privileges, the potential for system compromise is significant; if it runs under restricted rights, the damage may be limited to files within the allowance of those permissions.