vp-interaction-routing
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The bridge script
scripts/codex-cua-bridge.mjsspawns a local subprocess using thecodexbinary found within the officialChatGPT.appbundle on macOS. This is a functional requirement to interface with the application's experimental 'Computer Use' features. - [DYNAMIC_EXECUTION]: The skill uses a persistent Node.js REPL (
node_repl) provided by the ChatGPT app-server to execute JavaScript snippets that call accessibility functions. The script implements safe argument interpolation by using double-JSON stringification to prevent code injection into the JavaScript templates. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from web pages and application accessibility trees. It includes explicit instructions in
references/browser-routing.mdandreferences/native-ui-routing.mdto treat this external content as untrusted data rather than agent instructions, and to maintain user confirmation policies for all mutating actions. - [SAFE]: No malicious patterns such as data exfiltration to unauthorized domains, hardcoded credentials, or obfuscated payloads were detected. The script uses only built-in Node.js modules and targets official software from a well-known vendor.
Audit Metadata