Check a codebase for malicious code before you run it
Before installing or running an unfamiliar repo, have Jev answer yes/no risk questions about its install scripts and suspicious files.
The use case
Developers clone repositories and install packages they have never read, and supply-chain attacks hide in install scripts and obfuscated files. This scanner walks the code and asks Jev typed yes/no questions about the risky spots, such as whether a script sends data to an outside host or downloads and runs something. Flagged files are listed with the reason so a person can read them. It adds a cheap first look, not a guarantee: a developer still reviews anything flagged and should run unknown code in a sandbox or container.
Example workflow
- Clone the repository into a sandboxed folder
- List install scripts, post-install hooks and obfuscated files
- Ask Jev Noul questions on each about network calls and code execution
- Show flagged files with reasons in the terminal
- A developer reads flagged code and runs it only in a container
Tools
- Jev (TypeSafe AI)
- Docker
- GitHub
Risks and guardrails
- A clean result does not mean the code is safe
- Attackers can write code designed to look harmless
- Sending private code to an API may break policy
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- is-malicious by Paul on Made with Jev: madewithjev.com/builds/is-malicious
- is-malicious: original post or repo by Paul: github.com/luantak/is-malicious
- Docker MCP Catalog and Toolkit (official page): docs.docker.com/ai/mcp-catalog-and-toolkit (checked 2026-11-09)
- GitHub MCP server (official page): github.com/github/github-mcp-server (checked 2026-11-09)