Hermes’s browser automation is powerful — it can drive a real Chrome browser to navigate websites, fill forms, and extract data. But with great power comes great responsibility. What if the agent accidentally browses to a sensitive page?
The Fix
Two changes hardened the CDP browser integration:
-
Private-page guard applied to
browser_cdpframe_id routing — when the agent navigates to a page that requires authentication, the guard kicks in before any data is extracted. -
Camofox input blocked on private pages — the data extraction tool can’t run on pages flagged as private.
How It Works
The private-page guard works at the CDP frame level. When the agent uses browser_cdp to connect to a page, the guard inspects the page’s security context — checking for authentication headers, login forms, and known sensitive domains. If the page is flagged as private, the guard:
- Blocks tool calls that would extract data
- Prevents screenshots
- Stops the agent from filling forms or submitting data
- Logs the attempt for audit
Why This Matters
As coding agents gain the ability to browse the web autonomously, the security model needs to evolve. It’s not enough for the agent to be secure — the agent needs to keep your data secure too.
Hermes’s browser automation includes full CDP integration with a private-page guard. Oh My Pi has Puppeteer-based browsing with its own security model. The private-page guard in Hermes adds an extra layer by inspecting page security context before allowing data extraction.