deep-research
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a research methodology that fetches and processes untrusted content from the web via the
WebFetchandWebSearchtools. This creates an indirect prompt injection surface where malicious content on a researched website could attempt to influence the agent's behavior. - Ingestion points: The skill utilizes
webReader,firecrawl_scrape, andcurlto ingest data from arbitrary URLs during the research and monitoring phases. - Boundary markers: The instructions do not explicitly mandate the use of XML tags or specific "ignore embedded instructions" delimiters when the agent processes fetched text.
- Capability inventory: The skill utilizes the
Bash,Write, andEdittools, which represent the potential impact surface for an injection attack. - Sanitization: While the skill emphasizes manual source validation and triangulation (triangulation principle), it does not define automated sanitization or filtering of external content within the prompt context.
- [EXTERNAL_DOWNLOADS]: The skill facilitates data retrieval and tool integration from several well-known and authoritative security services.
- Fetches vulnerability data from the NIST National Vulnerability Database (NVD) API.
- Retrieves threat intelligence and technique mappings from the MITRE ATT&CK official GitHub repository.
- Accesses the CISA Known Exploited Vulnerabilities (KEV) catalog.
- Integrates with established security platforms including Shodan, VirusTotal, Censys, and AbuseIPDB via their respective APIs.
- [COMMAND_EXECUTION]: The skill methodology and payload templates extensively utilize the
Bashtool to perform research tasks using standard security utilities. - Employs
curl,jq, andgrepfor API data extraction and processing. - Uses security tools such as
searchsploit,nmap,dig,subfinder, andexiftoolfor intelligence gathering and metadata analysis. - Utilizes
npxto execute Model Context Protocol (MCP) servers for specialized intelligence enrichment.
Audit Metadata