walkinto

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

• Third-party content exposure detected (medium risk: 0.30). Helpdesk tickets read flow (tickets.js view <ticketId> → /agent-api/helpdesk/tickets/:id returns full ticket thread) ingests outsider-authored customer support text (ticket subjects/bodies/messages) into the runtime LLM without any prior selection step specific to an item beyond the attacker-controlled ticket content being fetched/read.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill uses the runtime-configured base URL (https://walkinto.in) to fetch Knowledge Base articles and other /agent-api responses that are then injected into assistant replies (e.g., KB markdown cited for ticket replies), so remote content from https://walkinto.in can directly influence the agent's prompts/output.