ai-model-wechat
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents a workflow for processing untrusted user inputs (text and images) and performing database operations, which represents an indirect injection surface.\n
- Ingestion points: Untrusted data enters the agent context through the
messagesarray ingenerateTextandstreamTextcalls, specifically via text andimage_urlblocks in the code examples inSKILL.md.\n - Boundary markers: The documentation recommends using structured prompt instructions to enforce specific output schemas (JSON) and to ignore extraneous model commentary.\n
- Capability inventory: The skill demonstrates capabilities for file storage via
wx.cloud.uploadFileand database persistence viadb.collection().add.\n - Sanitization: The instructions provide mitigations such as regex-based cleaning of AI output and strictly mandate human verification of extracted data before it is saved to a database.\n- [SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration behaviors were detected. All external links point to official vendor documentation and resource management domains. The skill also includes proactive warnings regarding deprecated or offline models.
Audit Metadata