chrome-bridge-automation
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: Executes browser automation tasks via
npx @midscene/web@1using theBashtool. - [EXTERNAL_DOWNLOADS]: Fetches the
@midscene/webpackage from the npm registry during execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes live web content through screenshots to drive its logic, which exposes the agent to malicious instructions potentially embedded in external websites. * Ingestion points: Screenshots of web pages processed by the agent as described in SKILL.md. * Boundary markers: No delimiters or boundary markers are defined to separate web content from instructions. * Capability inventory: Shell command execution and full browser interaction including navigation and form filling. * Sanitization: No sanitization or filtering of visual content is performed.
- [DATA_EXFILTRATION]: The 'Bridge mode' capability connects to the user's active Chrome session, granting the skill access to cookies, login states, and private data. This poses a risk of exposure if the agent is directed by untrusted web content to scrape or transmit this information.
Audit Metadata