sandcastle-runner

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

• Third-party content exposure detected (medium risk: 0.65). Runtime path sandcastle_runner.run_sandcastle() launches npx sandcastle run and streams the subprocess stdout line-by-line into session.logs after only regex redaction; Sandcastle (and any embedded prompts it uses) is not operating-user-authored, so its free-text output can include outsider-authored content that becomes LLM context via the dashboard/session log pipeline.