umodel-rca

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime workflow uses umctl/umodel-query to query a user-selected UModel workspace and then ingests free-text fields from stored entities (service names/incident IDs, and linked runbook/observations) via .entity, .entity_set | entity-call get_metrics, and .entity_set | entity-call get_logs, meaning outsider-authored content in that workspace can flow into the LLM without prior selection beyond the initial incident/service identification.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's Setup instructs runtime installation of remote code via "go install github.com/alibaba/UnifiedModel/cmd/umctl@latest", which fetches and builds/executes external code that the skill requires (github.com/alibaba/UnifiedModel/cmd/umctl@latest).