ads

Pass

Audited by Gen Agent Trust Hub on May 5, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface
  • Ingestion points: The skill ingests untrusted data from external websites via the /ads dna and /ads audit commands. These commands trigger scripts/fetch_page.py and scripts/analyze_landing.py to retrieve and parse content from user-provided URLs.
  • Boundary markers: There are no explicit boundary markers or instructions within the processing scripts or the brand DNA templates to protect against instructions embedded in fetched website content (e.g., in HTML comments, meta descriptions, or page body).
  • Capability inventory: The skill possesses extensive capabilities, including browser automation via playwright (scripts/capture_screenshot.py), outbound network communication to image generation APIs (scripts/generate_image.py), and the ability to write generated campaign assets and reports to the local file system.
  • Sanitization: While the extraction scripts structure data into JSON formats (e.g., brand-profile.json), the raw text extracted from titles, headers, and page content is eventually processed by the agent to define brand voice, values, and campaign strategy, which could be manipulated by adversarial website content.
Audit Metadata
Risk Level
SAFE
Analyzed
May 5, 2026, 10:42 AM
Security Audit — agent-trust-hub — ads