agents-sdk
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection Surface: The skill includes patterns for processing external data from sources like emails (
onEmail), webhooks (onRequest), and web scraping (browser_search). These ingestion points (noted inreferences/email.mdandreferences/webhooks-push.md) lack explicit boundary markers in the provided examples, though they utilize capabilities like SQL storage, network requests, and dynamic code execution. Implementing validation and sanitization for these inputs is recommended. - Dynamic Execution Features: The skill documents experimental features such as 'Codemode' and browser tools (
references/codemode.mdandreferences/browse-the-web.md). These allow for the execution of generated JavaScript in isolated Worker or browser environments to orchestrate complex tool chains. - Resource Management: The skill provides instructions for installing official packages (e.g.,
agents,@cloudflare/ai-chat) and referencing documentation from the vendor's official domains.
Audit Metadata