burner-agents-privacy

Warn

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the user to clone a repository from an unverified account ('NotPBShaw') and install dependencies via pip. This introduces a risk of executing unvetted code from an unknown third-party source.
  • [PROMPT_INJECTION]: The skill uses a 'Reasoning Engine' to interpret and act on live web page content. This is a classic surface for indirect prompt injection, where malicious instructions on a website could take control of the agent's actions.
  • Ingestion points: Browser context content is ingested via ReasoningEngine.reason_and_act and engine.analyze_page in SKILL.md.
  • Boundary markers: There are no boundary markers or explicit instructions provided to the reasoning engine to distinguish between system instructions and untrusted page content.
  • Capability inventory: The agent has the capability to navigate, interact with web pages, and generate research reports based on its findings.
  • Sanitization: The skill lacks evidence of content sanitization or validation before the live page data is processed by the underlying language model.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 21, 2026, 02:24 AM
Security Audit — agent-trust-hub — burner-agents-privacy