At a Glance
| Vendor | |
|---|---|
| Website | github.com/google-gemini/gemini-cli |
| Category | Coding Assistant |
| Default Risk Rating | Medium |
| Default Status for New Clients | Unapproved |
| Recognized By | Its client signature, so it has its own status even on domains it shares with another product |
| Desktop App Inventoried | Command Line (Client Signature) |
| Extension Inventoried | VS Code, Cursor and Windsurf |
| Coverage | Content Inspection |
| Catalog Reviewed | October 28, 2026 |
The risk rating and default status are ProctoredAI's starting points for a new client. Each MSP sets the status per client. How apps are rated and verified.
What to Know About Gemini CLI
Google's open-source coding agent for the terminal: reads and edits files and runs commands. Identified by its client signature (GeminiCLI user agent) when it runs with a Gemini API key; sign-in with a Google account uses a Google host that is not inspected yet.
Why Coding Assistant Tools Need Governing
Coding assistants read source code and often the secrets in it: API keys, connection strings and customer data in test fixtures.
Approve the assistants the client licenses, block credentials and secrets everywhere, and review the MCP servers developers connect them to.
How ProctoredAI Governs Gemini CLI
- See it. The agent reports every device and account using Gemini CLI, and finds its extension in the browser or code editor.
- Decide it. Mark it Approved, Unapproved or Blocked for each client. Unapproved use can ask the person for a business reason first.
- Protect what is sent. Prompts and uploads are inspected on the device, and credentials, health and payment data and sensitive documents are stopped, redacted or warned on by the client's rules.
- Prove it. Usage, decisions and blocked events land in the client's monthly report as evidence.
Policy Packs That Fit
- Credentials & Secrets. Block API keys, private keys, passwords and connection strings everywhere.
- Prompt Injection & Jailbreaks. Block prompts that try to override the model's instructions or extract its system prompt, and uploaded content that carries hidden directives for the AI.