← Back to Intel Center

2026-11-16 Daily Security Intelligence

RSS

CSSA Exclusive Early Warning 2

CSSA Exclusive Early Warning 10.0 Critical

Tool-result injection of slash commands in oh-my-openagent lets attackers hijack session state

An attacker can exploit a trust-boundary flaw in the native subscription-bridging mechanism when it handles tool return results. By placing a special slash-prefixed instruction such as `/goal` at the start of a text block, the attacker induces the backend CLI to misidentify it as a user-level control command. This behavior directly registers a session-level Stop hook and changes the active goal state, bypassing normal permission checks and data-isolation mechanisms, causing unexpected session interruption or logic hijacking, and seriously damaging system integrity and availability.

Component
oh-my-openagent is an agent-orchestration framework built on the Senpi engine and the Claude Agent SDK. Its core architecture depends on an Anthropic subscription-bridging module to enable multi-agent collaboration. The component is designed to convert protocols in a standardized way so that tool-execution results from the host environment are passed seamlessly to the large-language-model interface, supporting complex task decomposition, context synchronization, and automated workflow orchestration, and keeping data flow and logical closure across services intact.
Type
Injection (CWE-74)
Repo
Remediation
  • Apply strict input sanitization and format validation at the data-transfer layer, and treat tool-output content as plain text only. Establish a clear trust boundary so that unverified raw data is not parsed directly as an internal control instruction, and escape or filter specific prefix characters, cutting off the illicit path from data to command at the root.
CSSA Exclusive Early Warning 10.0 Critical

Flaws in Director's CodeExecutor and ChatNamespace allow remote unauthorized code injection and loss of system control

An attacker can exploit the lack of connection authentication in ChatNamespace and its loose cross-origin resource sharing policy, sending unauthenticated requests directly to the server over the network and causing CodeExecutor to run malicious Python code generated by an LLM. Because the execution environment exposes the full built-in function library, the attacker can bypass sandbox restrictions, call operating-system interfaces directly, and run arbitrary commands, fully controlling the server and seriously damaging confidentiality, integrity, and availability.

Component
Director is a Python-based video analysis and management platform. Its core architecture includes a CodeExecutor component that handles model-generated timeline code and a Socket.IO ChatNamespace module for real-time communication, and it is designed to provide automated video-content processing, interactive session management, and distributed task scheduling.
Type
Code Injection (CWE-94)
Repo
Remediation
  • Disable dangerous built-in functions such as os and sys in the code-execution environment, and apply a strict allowlist to limit importable modules. Also enforce authentication-token checks at the WebSocket connection layer, and run the workload in a least-privilege container to isolate potential system-level risk.

CVE Intelligence 2

CVE-2026-129449 CVSS 9.9 Critical 2026-11-15

Command injection in DrayTek Vigor2960 leads to remote code execution

In DrayTek Vigor2960 version 1.5.1.6 and earlier, the sub_14914 function in `/cgi-bin/mainfunction.cgi` mishandles the swmconfigupload parameter, a command-injection flaw (CWE-74). An attacker can remotely manipulate this parameter and run arbitrary system commands on the target, threatening the device's confidentiality, integrity, and availability. The exploit method has been publicly disclosed, and the attack can be launched remotely with no user interaction. Every DrayTek Vigor2960 device that has not been upgraded to version 1.5.1.7 is affected. The attacker only needs to send a crafted malicious request to the target device and may gain complete system control.

Component
DrayTek Vigor2960 is an enterprise-grade broadband router/firewall used for network access and security control.
Risks
  • Complete system control: The attacker can use this flaw to run arbitrary commands on the victim device and, depending on system privileges, install programs, view, change, or delete data, or create new accounts with full privileges
  • Attack without user interaction: The attacker can trigger the flaw with no user interaction by launching the attack directly over the network
  • High-impact damage: Because the CVSS score is extremely high, a successful exploit may fully take over the device and then use it as a pivot to attack other assets on the internal network
Source
Remediation
  • Upgrade DrayTek Vigor2960 immediately to version 1.5.1.7 or later
  • Monitor access logs for `/cgi-bin/mainfunction.cgi` and identify abnormal requests
  • Deploy firewall rules at the network boundary to restrict unauthorized access to the management interface
CVE-2026-129453 CVSS 9.9 Critical 2026-11-15

Heap overflow in DrayTek Vigor2960 leads to remote code execution

In DrayTek Vigor2960 version 1.5.1.6, the check_file_exist function in `/cgi-bin/mainfunction.cgi` of the File Existence Check component has a security flaw. Manipulating the upload_name parameter can trigger a heap-based buffer overflow (CWE-119). The flaw allows an attacker to launch the attack remotely, exploit code has been made public, and it may lead to remote code execution or interruption of system services. Every DrayTek Vigor2960 device running version 1.5.1.6 or earlier is affected. The attacker can trigger the flaw remotely without complex interaction; the exploit is easy and the damage is severe.

Component
DrayTek Vigor2960 is an enterprise-grade network-access device that provides broadband connectivity, firewall, and remote-access functions.
Risks
  • Complete system control: The attacker can use this flaw to run arbitrary code on the device and obtain the highest level of control
  • Attack without user interaction: The attack vector is launched over the network, complexity is low, and the attacker can exploit it with no user interaction
  • Data leakage and tampering: Because the impact on confidentiality, integrity, and availability is high in every case, the attacker can steal sensitive data or disrupt normal device operation
Source
Remediation
  • Upgrade immediately to version 1.5.1.7 or later to fix this flaw
  • Monitor network traffic for abnormal requests targeting `/cgi-bin/mainfunction.cgi`
  • Restrict access to the management interface and allow connections only from trusted IP addresses