ai-runtime-security-sandbox

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.70). The GitHub clone URL is hosted under an individual/unfamiliar account and could be used to distribute code or binaries without an established reputation, so it represents a higher-risk download/source to verify before executing.

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

• Third-party content exposure detected (medium risk: 0.30). Indirect prompt injection is driven by user-supplied chat input sent to /api/chat (message field) which the runtime then uses for RAG retrieval against the local data/documents/*.md knowledge base.

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

• Potentially malicious external URL detected (high risk: 0.90). The README instructs cloning and running code from https://github.com/TatarinBlack/ai-runtime-security-sandbox.git (git clone ... then python run.py), which fetches remote code and executes it locally, so the URL is a runtime external dependency that results in executing fetched code.