page-visuals

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly asks for and passes a public source URL and arbitrary http(s) refs to the Bloom provider (see SKILL.md "onboard the brand from the source URL" and scripts/providers/bloom.sh which posts /brands and uploads remote refs), causing the system to fetch and ingest untrusted third‑party web content that is then used as brand context influencing prompt generation and provider/QA decisions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's runtime provider adapter (scripts/providers/bloom.sh) sends an operator-supplied --source-url to Bloom's API (https://www.trybloom.ai/api/v1, e.g. POST /brands) to onboard brand content which Bloom will fetch and use as brand context that directly influences image generation, so external URL-hosted content can control the provider's prompts/results.