security-detections-mcp

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime ingests detection rule text from user-configured local directories populated via public third-party repos (e.g., SigmaHQ/sigma, splunk/security_content, elastic/detection-rules, etc.), so the LLM context can include outsider-authored free-form rule bodies when tools like search, get_by_id, or list_all return those texts.

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

• Potentially malicious external URL detected (high risk: 0.90). The hosted MCP endpoint https://detect.michaelhaag.org/api/mcp/mcp is explicitly configured as a remote MCP server used at runtime by the client (via claude mcp / mcp-remote), meaning responses from that URL can control available tools/prompts and thus directly influence agent behavior.