Search Realtime Information-hardened
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill contains defensive instructions designed to prevent the agent from being manipulated by chat-injected directives claiming administrative or developer authority. These instructions strengthen the model's adherence to the tool-first search requirement.
- [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted data from user input and external search results.
- Ingestion points: The {{topic}} variable in SKILL.md and the results returned by the web_search_preview tool.
- Boundary markers: Absent; user input is directly interpolated into the search prompt template.
- Capability inventory: The skill uses the web_search_preview tool to access the live web.
- Sanitization: No sanitization or escaping is applied to the input or the retrieved search results before processing.
- [DATA_EXFILTRATION]: Analysis of the skill instructions and documentation shows no attempts to access sensitive local files or exfiltrate data to unauthorized domains. All external references are limited to the vendor's official resources.
- [NO_CODE]: The skill is implemented purely through markdown instructions and tool definitions, containing no executable scripts or third-party code dependencies.
Audit Metadata