keyword-deep-dive
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions require the agent to search for keywords and read the full content of the top-ranking results from the web, creating a surface for indirect prompt injection. A malicious actor could optimize a page to rank for a target keyword and include hidden instructions designed to hijack the agent's behavior when the page is processed.
- Ingestion points: Found in
SKILL.mdunder Step 1 ('Read the top 10 results in detail') and Step 5 ('Fetch and read the top 3 pages'). - Boundary markers: Absent. The instructions do not define delimiters or provide 'ignore instructions' warnings for the external content being processed.
- Capability inventory: The agent uses web search and browsing tools to retrieve external data and produces detailed content strategy reports.
- Sanitization: Absent. There is no mechanism described to filter or sanitize the retrieved external content before it is processed by the agent.
Audit Metadata