marketing-pricing
Pass
Audited by Gen Agent Trust Hub on Apr 27, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires and provides instructions to install the
agent-browserdependency from Vercel Labs' official GitHub repository. - [COMMAND_EXECUTION]: It utilizes shell commands to perform automated browser research, capture screenshots, and extract text from external websites.
- [PROMPT_INJECTION]: The skill is subject to indirect prompt injection due to its ingestion of untrusted external data during research activities.
- Ingestion points: Accesses and parses content from competitor websites, G2 reviews, Reddit posts, and the Wayback Machine via
agent-browser. - Capability inventory: Performs network requests to various domains and writes research deliverables (markdown and images) to the local brand workspace.
- Boundary markers: No explicit delimiters or specific 'ignore' instructions are used to isolate untrusted web content from the agent's core instructions.
- Sanitization: The skill lacks mechanisms to sanitize or validate scraped text to prevent embedded instructions from influencing the agent's behavior.
Audit Metadata