cnki-advanced-search
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted search definitions which exposes it to potential indirect instruction execution:
- Ingestion points: Natural language search criteria enter the context via the
$ARGUMENTSvariable inSKILL.md. - Boundary markers: Absent. No boundary delimiters or filtering rules are declared to separate potential adversarial prompts from benign search queries.
- Capability inventory: The agent has capabilities to navigate pages and execute arbitrary scripts via browser developer tools (
mcp__chrome-devtools__navigate_pageand script evaluation). - Sanitization: Absent. There are no guidelines instructing the model to validate or escape quotes and special characters within the user input.
- [DYNAMIC_EXECUTION]: The skill involves script generation where user-supplied content is interpolated directly into a template script string (e.g., substituting text into
const query = "KEYWORDS";). This dynamically assembled script block is then executed as code inside the browser session, introducing a risk of broken syntax or execution anomalies if the text contains unescaped string delimiters.
Audit Metadata