illustration-ideas

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [External Downloads] (LOW): The skill instructs the agent to suggest 'npm install -g agent-browser' to the user if a browser tool is missing. This introduces a dependency on a third-party package not included in the trusted sources list.
  • [Indirect Prompt Injection] (LOW): The skill processes content from arbitrary URLs, which may contain malicious instructions designed to manipulate the agent's behavior.
  • Ingestion points: External data is ingested through the 'agent-browser' CLI or MCP browser tools (e.g., agent-browser open <url>).
  • Boundary markers: Absent. The skill lacks delimiters or instructions to ignore embedded commands within the fetched article content.
  • Capability inventory: The agent is restricted to browser-based data extraction and text/diagram generation (Mermaid/ASCII). No high-risk capabilities like file system modification or network exfiltration of local secrets were detected.
  • Sanitization: Absent. The agent is explicitly told to 'Carefully review the full content,' making it susceptible to following instructions found within the processed text.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:08 PM