deep-research
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches content from external web pages, making it susceptible to indirect prompt injection if web results contain malicious instructions targeting the agent.
- Ingestion points: Content is ingested from the web in
scripts/research.pywithin thesearch_single_keywordfunction using the DuckDuckGo Search API. - Boundary markers: No specific delimiters or boundary text (such as XML tags or explicit system instructions) are used to isolate untrusted text inside the generated Markdown report.
- Capability inventory: The execution environment allows powerful tools such as
Bash,Read,Write, andEditas defined inSKILL.md's allowed-tools configuration. - Sanitization: The
titleandbodyfields extracted from web search results are interpolated directly into the final Markdown document without any sanitization, filtering, or escaping.
Audit Metadata