batch-download
Warn
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill instructs the agent to use Python scripts to programmatically generate URL lists based on patterns discovered from web content. Generating and executing code derived from external patterns introduces a potential risk of script injection if the pattern extraction is not strictly sanitized.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from web discovery and search results to influence its planning and sub-agent delegation, creating an attack surface for indirect prompt injection.
- Ingestion points: Web page content, search results, and file metadata gathered during the discovery phase (SKILL.md).
- Boundary markers: Mandates strict output validation, specific JSON/Markdown schemas, and explicit sub-agent constraints (Phase 3 and 4).
- Capability inventory: Web discovery, browser tools, Python-based network libraries (requests, httpx), and file system access to /mnt/agents/output/ (Phase 2 and 3).
- Sanitization: Instructs verification of URL existence, file type matching, and content sanity checks (Phase 2 and 4).
- [COMMAND_EXECUTION]: The skill employs various tools and libraries, including Python, requests, and BeautifulSoup, to perform network operations and parse content. These capabilities represent a standard attack surface when interacting with external data.
Audit Metadata