cnki-journal-search

Warn

Audited by Gen Agent Trust Hub on Mar 3, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill's instructions for executing browser scripts are vulnerable to script injection.
  • Evidence: Step 2 uses a JavaScript template const query = "QUERY_HERE"; and instructs the agent to replace the placeholder with the search term.
  • Risk: A crafted search term containing quotes and semicolons could break the string literal and execute arbitrary JavaScript in the browser context where the tool runs.
  • [EXTERNAL_DOWNLOADS]: The skill performs automated navigation to an external service.
  • Evidence: Step 1 uses mcp__chrome-devtools__navigate_page to access https://navi.cnki.net/knavi.
  • Context: This is a legitimate interaction with the CNKI academic database for the purpose of the skill.
  • [DATA_EXFILTRATION]: The skill extracts and returns content from the target website.
  • Evidence: The script in Step 2 parses the DOM and returns journal metadata including names, ISSNs, and impact factors to the agent.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection from the web content it processes.
  • Ingestion points: Web content is extracted via document.body.innerText and querySelectorAll in Step 2.
  • Boundary markers: No delimiters or boundary markers are used to isolate the untrusted data.
  • Capability inventory: The skill has the capability to navigate pages and execute browser scripts via Chrome DevTools.
  • Sanitization: No sanitization or validation is performed on the extracted journal data before it is returned to the agent context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 3, 2026, 04:57 PM