Coding agents are installing untrusted software on your endpoints. No one is reviewing it.
Skills, MCP servers, plugins, and extensions pull in from GitHub and public registries, running with developer credentials, spawning processes, and reading source code and secrets. Dizzy gives your security team the visibility and controls to catch a problem before it becomes a breach.
Skills, MCP servers, plugins, and shell commands run with developer-level access, with no security review or visibility.
Claude Code, Cursor, GitHub Copilot, Codex, and similar tools extend themselves at runtime. Each extension is an untrusted code execution path your security team has no visibility into.
Untrusted supply chain
Skills and MCP servers pull in from GitHub and public registries with no ownership verification, no signing, and no security review. Packages can be swapped out after your team approved them.
High-privilege access by default
Extensions run with the developer's credentials. They can read source code, credentials, and cloud configs, execute shell commands, spawn processes, and send data to external hosts.
Zero security visibility
Skills and extensions can be installed and running before your security team knows they exist. Malicious behavior is often time-delayed or condition-triggered, so static review won't catch it.
Security built for the coding agent attack surface.
Dizzy covers the skills, MCP servers, plugins, and extensions that existing security tools were never designed to handle.
See what is running
Know every agent, extension, and tool active across your developer endpoints.
Understand the risk
Surface threats that file scanning and reputation checks are blind to.
Apply control
Set policy and act on findings without getting in your engineers' way.
Full visibility across every coding agent your developers run.
One place to see what agents are running, what they found, and what to do about it, across Claude Code, Cursor, Copilot, and more.
From the field.
Real attacks, disclosed responsibly. No target names. Just the technique and what it means for your team.
Your developers aren't waiting for security approval to install the next skill.
Talk to us about what's running on your endpoints today.