manage-data

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill prompt instructs the user to include an API key directly in a Git clone URL (git clone https://<user>:<api-key>@git.llm-stats.com/...), which encourages insecure handling of secrets by exposing them in command history and potentially in agent logs.

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

• Third-party content exposure detected (medium risk: 0.30). The workflow involves pulling and managing benchmark datasets via the ZeroEval SDK and CLI, which ingests benchmark row data (e.g., questions and answers) from the ZeroEval backend. Because benchmark content can originate from external or untrusted users contributing to shared datasets, this represents an indirect prompt injection risk through ingested data rows.