agentpmt-audit-logs

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

• Third-party content exposure detected (medium risk: 0.65). This skill’s required runtime workflow can call get_chat_review / get_workflow_run / list actions that return chat transcripts and workflow results (outsider-authored free text could be included in those stored conversations/messages), which the agent then ingests into the LLM context as returned JSON content.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly calls the AgentPMT MCP/REST endpoints at runtime to fetch live schema and instructions (which it treats as authoritative), specifically https://api.agentpmt.com/mcp/ and https://api.agentpmt.com/products/purchase, so remote responses can directly control prompts/behavior.