See how Qpoint secures the hidden agent surface.Read the blog post →

Policy demo
Control the hidden agent surface

Agents write and run scripts, and the agent harness can’t see which files those scripts read or where they send requests. Qpoint runs underneath the agent, so its policies apply to what those scripts do. In the demo, one rule flags any read of a secrets file, catching Claude Code with curl and again with a Python script, then blocks the read when the rule is switched to deny.
Gaurav Mahajan
Gaurav Mahajan
November 9, 2026

About this demo

Most agent security stops at the tool call. But agents also write and run scripts. The agent harness can’t see which files those scripts open or where they send requests. That’s the hidden agent surface. Qpoint runs underneath the agent, so its policies also apply to what those scripts do.

In this demo, one Qpoint policy flags any read of a project’s .env file, which holds a GitHub access token. We ask Claude Code to grab the token and list open PRs with curl. It’s one tool call, but the shell reads the .env file as part of it. Qpoint flags that read and ties it back to the session. We then ask Claude to do the same thing with a Python script. Qpoint flags the script’s read of the file too.

Next, we switch the rule from flag to deny and run the same prompt again. The read fails, and the token stays on the machine. The policy never mentions curl or Python. It only covers the file, and Qpoint enforces it no matter how the agent tries to read it.

See Qpoint in action in your environment
Request access and we'll get you set up.
Request Access