sdframe

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

• Third-party content exposure detected (high risk: 0.78). SKILL.md’s required runtime workflow (Mode A/B) instructs the agent to read outsider-authored public reference content fetched from sdframe.caldis.me (e.g., references/catalog.md, references/frameworks/{slug}.md, and llms-full.txt), which is free-form text that can contain indirect prompt-injection payloads.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs agents to fetch runtime machine-readable agent context and framework records from public endpoints (e.g., https://sdframe.caldis.me/llms-full.txt and related API endpoints) which are used to control agent prompts and selection/apply flows, so these external URLs are required runtime dependencies that directly influence agent instructions.