Submit #886431: Simular AI Agent-S 0.3.2 Denial of Service

Agent-S CodeAgent contains a denial of service vulnerability in its local tool execution workflow. When local code execution is enabled, a crafted task prompt can induce CodeAgent to execute model-generated Python code. The Python subprocess is run without an effective timeout, memory budget, or stdout/stderr output-size limit, and its output is fully buffered before being returned to the agent.
An attacker can cause high-volume output or memory-consuming Python execution through a normal task prompt, exhausting the worker process memory and terminating the agent service. In a deployed agent environment, this can interrupt normal task execution and make the service unavailable to legitimate users.

• Source: ⚠️ https://github.com/outlookgp/CVE/tree/main/AgentS_CodeAgent_Tool_Output_DoS_Report
• User: outlookgp (UID 99658)
• Submission: 07/11/2026 13:32 (2 months ago)
• Moderation: 09/02/2026 16:16 (2 months later)
• Status: Accepted
• VulDB entry: 398133 [simular-ai Agent-S 0.3.1/0.3.2 CodeAgent code_agent.py denial of service]
• Points: 20