efficient-web-research
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes processes for ingesting data from external web sources and the GitHub API, which presents a surface for indirect prompt injection.
- Ingestion points: The skill instructs the agent to use
read_url_content,browser_subagent, andsearch_webto retrieve data from arbitrary URLs and search results (SKILL.md). - Boundary markers: While the skill suggests summarizing content and stripping boilerplate to isolate data, it does not explicitly mandate the use of XML-style tags or delimiters to separate untrusted content from system instructions.
- Capability inventory: The skill relies on tools capable of reading remote file content and browsing the web.
- Sanitization: The protocol includes specific instructions to strip non-content elements such as navigation menus, cookie banners, and ads from fetched HTML (SKILL.md).
Audit Metadata