Before you hand an AI agent the keys to your terminal, you should probably know what's in its reach. That's the premise behind agentcheck, a new open-source CLI tool that scans your local environment and reports every credential, API key, and access point an AI agent could touch.
The tool runs read-only, meaning it doesn't modify anything. It checks for AWS, GCP, and Azure IAM credentials (the identity systems that control access to cloud resources), over 100 common API key environment variables, Kubernetes cluster contexts, Docker configurations, SSH keys, Terraform state files, and .env files where developers typically store secrets. Each finding gets tagged with a severity level from LOW to CRITICAL, so you can focus on what actually matters rather than sorting through noise.
This fills a real gap. Right now, most people running AI coding agents like Cursor, Claude Code, or Aider are giving those tools access to their full shell environment by default. That means every credential stored in an environment variable, every cloud login cached on disk, every SSH key sitting in ~/.ssh is technically reachable. Most developers have never audited what that actually includes, and the answer is usually more than they expect.
The practical use case is straightforward: run agentcheck before you start an AI agent session, see what's exposed, and decide if you need to revoke anything or run the agent in a sandboxed environment. It's the kind of tool that takes 30 seconds to run and could save you from a serious credential leak.
One limitation: agentcheck tells you what's accessible, not what an agent will actually try to use. A well-behaved agent won't go fishing through your AWS credentials, but the point is that it could. As AI agents gain more autonomy and tool-use capabilities, knowing your exposure surface before granting shell access is basic hygiene that most people are skipping entirely.