claude-office-skills: why it is not safe to use
We found no hidden data exfiltration or credential access, but the skill can overwrite user-chosen files and the README suggests piping a remote installer to bash.
What we found
Our review did not find hidden instructions, hardcoded secrets, credential collection, privilege escalation, or undisclosed data exfiltration. The sampled skill content and MCP server behavior appear focused on processing user-provided local files and returning results.
The warning is based on two narrower security findings. First, the office-mcp tooling can write output files such as merged, compressed, watermarked, split, or filled documents to user-specified paths, and we did not find a built-in confirmation step at the tool layer before an existing file is overwritten. That raises the chance of accidental file replacement if a path is reused. Second, the README promotes a one-line installation method that pipes a remote `install.sh` script directly into `bash`. Even when the source location is disclosed, this remote fetch-and-execute pattern reduces the chance to inspect what will run first.
What to do instead
Use this skill with explicit output paths and check whether destination files already exist before running write operations. Prefer reviewing installation scripts first, or install from cloned source or pinned release artifacts instead of piping a remote script into `bash`. This helps reduce avoidable file overwrite and install-chain risk.
Source: https://github.com/claude-office-skills/skills
We report what our security review found at the time we checked, with the goal of keeping people safe. Projects change; if a maintainer has since fixed this, we are glad to recheck it. Email hello@agentpod.com.