coze-web-search
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill depends on the
coze-coding-dev-sdklibrary to interact with the Coze platform. This is a functional dependency for searching and summarization capabilities. - [PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it processes untrusted content from the web and AI-generated summaries. * Ingestion points: External data is retrieved via
client.webSearch,client.advancedSearch, andclient.imageSearchinscripts/search.ts. * Boundary markers: The results are formatted using Markdown and text headers (e.g.,## Results,AI SUMMARY), but the skill does not provide explicit instructions to the agent to disregard any commands embedded within the fetched data. * Capability inventory: The script executes network requests to the Coze API. The overall risk depends on other tools available to the agent in its execution environment. * Sanitization: Content is truncated for length in the output functions, but there is no semantic filtering or sanitization to prevent the agent from interpreting instructions hidden in the search results.
Audit Metadata