humanrail-escalation

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow ingests outsider-authored free text because Workflow 3 explicitly “Pull[s] the raw incident from ServiceNow” and places the “Description: [full description, sanitized]” into the HumanRail task payload sent to external workers (workers handle arbitrary incident text submitted by users).

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 HumanRail MCP endpoint https://humanrail.dev/mcp (and the API base https://api.humanrail.dev/v1) to receive human worker outputs that directly determine whether the agent executes actions, so these external URLs are runtime dependencies that control agent instructions.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill's API explicitly includes payment-related parameters and billing controls: create_task accepts payout_currency and payout_max_amount, example workflows set payouts, and get_usage returns billing summary. Those fields are specific, domain-level controls to instruct/pay workers (i.e., move money) rather than generic tooling. This grants the agent direct financial execution authority (ability to initiate/pay tasks and view billing).