material-3

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md requires refreshing from a live catalog via curl (public web content) which would ingest outsider-authored free text into the agent’s LLM context through the downloaded design.md/catalog data path at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The SKILL.md explicitly instructs the agent to refresh runtime instructions by curling https://api.agent-ds.oday-bakkour.com/v1/systems/material-3/design.md, which would fetch remote content (design.md) that directly controls prompts/instructions.