alibabacloud-flink-knowledge

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). Layer 2 mandates WebFetch of Alibaba Cloud help pages at runtime (“Mandatory WebFetch … invoke WebFetch on a relevant help.aliyun.com/zh/flink/* URL”), and fetched page body text is then used as evidence in the agent’s LLM context, which is outsider-authored free text.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill mandates a runtime WebFetch to official docs (e.g. https://help.aliyun.com/zh/flink/*) which the agent must call and whose returned content directly controls the TOOL_EVIDENCE pre-output marker and downstream instruction/response logic, making it a required external dependency that influences prompts.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)