ah-fix-ui-bug
Pass
Audited by Gen Agent Trust Hub on Mar 27, 2026
Risk Level: SAFE
Full Analysis
- [DYNAMIC_EXECUTION]: The skill employs
chrome-devtools evaluate_scriptto inject JavaScript diagnostic tools into the browser context. These tools are loaded from local script files (e.g.,position-tracking.js,animation-logging.js) and customized by the agent to target specific elements. This includes monkey-patching standard browser APIs likeHTMLElement.prototype.animatefor logging purposes. - [COMMAND_EXECUTION]: Browser automation is achieved through the
chrome-devtoolsCLI, which executes commands for navigation, taking screenshots, and recording performance traces. This involves writing data to temporary local paths like/tmp/trace.json. - [INDIRECT_PROMPT_INJECTION]: The skill interacts with potentially untrusted web content and console logs, creating a surface for indirect prompt injection.
- Ingestion points: Web page content, a11y trees, and browser console messages accessed via the devtools protocol.
- Boundary markers: Not present.
- Capability inventory: Browser control (clicks, navigation), file system writes (screenshots, traces), and execution of JavaScript in the page context.
- Sanitization: Not explicitly implemented for the data retrieved from the browser before processing by the agent.
Audit Metadata