hsb-test

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

• Third-party content exposure detected (medium risk: 0.65). The skill ingests the user-provided test plan document at runtime (Phase 1, Step 1), and if that document is fetched from a public URL or is authored by an outsider, its free-form text is read and then incorporated into the agent’s LLM context for parsing/execution planning.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill will fetch a user-provided test-plan URL at runtime (e.g., "https://confluence.example.com/hsb-qa" as shown in Phase 1) via the WebFetch tool and then parse/run the test steps/commands from that document, so a malicious remote test plan could directly control executed commands inside the demo container.