Flaw in Claude Document Skills Allows Code Execution on Linux via LibreOffice
Researchers at AIR Security found a vulnerability named Office Party in Anthropic's official DOCX, PPTX, and XLSX skills for Claude. A helper script that launches LibreOffice on Linux loads a shared library from a fixed temporary path without verifying it, so an attacker-planted file can run with the privileges of the user operating the AI agent. The flaw is triggered when the workaround is used, which happens in environments where Unix-domain sockets are blocked, including some container setups and Claude Code's sandbox. Anthropic was notified in May and acknowledged the report but considers the issue outside its security boundary because the code runs inside a sandbox, and the helper script has reportedly remained unchanged. The researchers also showed that a separate, seemingly legitimate skill could write the file into the shared directory, meaning a compromised skill could chain into the vulnerable component. macOS is reported to be unaffected.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in