marketing-content
Pass
Audited by Gen Agent Trust Hub on Apr 27, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches and installs the
agent-browserskill from Vercel Labs' official GitHub repository to facilitate market research. - [COMMAND_EXECUTION]: Executes shell commands to install dependencies (
npm install) and conduct automated research sessions (e.g., Google PAA, Reddit threads) via theagent-browserCLI. - [PROMPT_INJECTION]: The skill maintains an indirect prompt injection surface by ingesting untrusted data from external websites during the content intelligence phase. 1. Ingestion points: Extracted body text from external research URLs via
agent-browser get text body. 2. Boundary markers: Absent. 3. Capability inventory: Subprocess execution viaagent-browserand file writes to the./brands/workspace. 4. Sanitization: No explicit sanitization or filtering of the fetched external text is described.
Audit Metadata