cnki-journal-search
Warn
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The search term provided via
$ARGUMENTSis directly interpolated into a JavaScript string literal within theevaluate_scriptblock:const query = "QUERY_HERE";. A malicious search term (e.g.,\"; alert(document.cookie); //) could break out of the string and execute arbitrary JavaScript code within the browser context of the CNKI website. - [INDIRECT_PROMPT_INJECTION]: The skill establishes an unsafe pattern for processing untrusted data by failing to sanitize or escape input before passing it to an execution environment.
- Ingestion points: The
$ARGUMENTSvariable is the primary source of untrusted data inSKILL.md. - Boundary markers: None. The input is blindly replaced into a template string.
- Capability inventory: Uses
mcp__chrome-devtools__evaluate_scriptwhich allows for full control over the browser's DOM and network requests in the current origin. - Sanitization: None present. The script assumes the input is a valid string literal.
Audit Metadata