web-research
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external websites and third-party documentation, which constitutes an indirect prompt injection attack surface.
- Ingestion points: Web content and technical documentation are fetched via MCP servers (Context7, Exa, Tavily) or built-in WebSearch/WebFetch tools (
SKILL.md). - Boundary markers: The skill includes explicit instructions to treat all external content as "untrusted data" and refers to a central discipline (
rules/untrusted-text.md). It also mandates that instructions within external documents should not be executed (references/source-evidence.md). - Capability inventory: Capabilities include information synthesis, summarization, and reporting. No direct file-system writes or arbitrary command execution are specified in the provided scripts.
- Sanitization: The skill implements "citation encoding" and "defensive framing" strategies to mitigate the risk of malicious instructions within processed text.
- [SAFE]: The skill provides explicit safety boundaries for high-risk operations, such as accessing the user's logged-in browser sessions. It mandates following tool-specific safety guidelines, prohibits the automatic bypass of MFA or payment systems, and requires that external output actions only occur upon direct user request.
Audit Metadata