story-long-scan

Warn

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSOBFUSCATION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/cdp-utils.js file uses child_process.execFileSync to execute the agent-browser CLI tool. It contains logic to dynamically resolve the executable path on Windows by searching the system PATH and reading npm shim files (agent-browser.cmd).
  • [DYNAMIC_EXECUTION]: The skill utilizes eval functions within a browser context (via the agent-browser tool) to extract information from web pages. The scraping logic is defined as JavaScript strings in scripts like scripts/qidian-rank-scraper.js and scripts/fanqie-rank-scraper.js and executed at runtime.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external web novel platforms, including book descriptions and tags written by third-party authors. This data is interpolated into the agent's context for analysis without rigorous sanitization, creating a surface where malicious instructions embedded in novel data could influence the agent's behavior.
  • [EXTERNAL_DOWNLOADS]: The scraper scripts fetch content from multiple external domains including qidian.com, fanqienovel.com, qimao.com, jjwxc.net, and ciweimao.com. These are well-known novel platforms used as the data source for the skill's primary function.
  • [OBFUSCATION]: The scripts/cdp-utils.js script employs Base64 encoding to pass JavaScript scraping logic to the agent-browser CLI tool. While intended to ensure argument integrity on Windows, this hides the executed scripts from simple string-based scanners.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 7, 2026, 01:44 PM