Executive Summary

CVE-2026-36045 is a high-severity OS command injection vulnerability in picoclaw versions 0.1.2 and earlier. It arises because the guardCommand() function, which is supposed to restrict shell command execution, uses an incomplete denylist of only 8 specific dangerous commands. This denylist fails to block many other harmful commands and techniques, allowing attackers to bypass it and execute arbitrary operating system commands on the host.

The vulnerability affects the ExecTool component in the pkg/tools/shell.go file. Since the denylist only blocks a narrow set of disk-destruction commands (like rm -rf, dd, shutdown), attackers can use other methods such as shell invocation (bash -c, sh -c), privilege escalation (sudo), scripting languages (python3, perl), and remote download-and-execute commands (curl | sh) to exploit the system.

Because the framework lacks an allowlist and executes any command not matching the denylist, this architectural flaw leads to multiple confirmed bypass payloads and enables attackers with access to the agent interface (e.g., messaging platforms like Telegram or Discord) to run arbitrary commands.

Useful 0 Not Useful 0

Impact Analysis

This vulnerability can have severe impacts if exploited. Attackers who gain access to the picoclaw agent interface can execute arbitrary OS commands on the host system.

• Full device control if picoclaw runs with root privileges, which is common in embedded deployments.
• Physical hardware manipulation through built-in I2C and SPI tools on IoT devices.
• Execution of remote payloads, reverse shells, and installation of persistence mechanisms.
• Potential credential theft and unauthorized access to sensitive information.

Useful 0 Not Useful 0

Mitigation Strategies

The vulnerability arises from an incomplete denylist in the guardCommand() function, allowing arbitrary OS command execution if an attacker has access to the agent interface.

Since there is no patched version available as of the advisory date, immediate mitigation steps include restricting or disabling access to the picoclaw agent interface (e.g., messaging platforms like Telegram or Discord) to prevent unauthorized command execution.

Additionally, running picoclaw with the least privileges possible (not as root) can limit the impact of exploitation.

Monitoring for suspicious command execution patterns or unusual activity on devices running vulnerable versions may also help in early detection.

Useful 0 Not Useful 0

Detection Guidance

This vulnerability involves OS command injection via the ExecTool component in picoclaw versions v0.1.2 and earlier. Detection involves monitoring for suspicious command execution patterns that bypass the incomplete denylist in the guardCommand() function.

Since the vulnerability allows arbitrary OS commands through interfaces like messaging platforms (Telegram, Discord), detection can focus on identifying unusual or unauthorized command executions or network activity related to reverse shells or remote payload downloads.

Suggested commands to detect exploitation attempts or suspicious activity include:

• Use process monitoring tools (e.g., `ps aux`, `top`) to look for unexpected shell or scripting language processes such as `bash -c`, `sh -c`, `python3`, or `perl`.
• Check for network connections that may indicate reverse shells or remote command execution using commands like `netstat -tulnp` or `ss -tulnp`.
• Search system logs for unusual command executions or errors related to ExecTool or shell commands.
• Monitor for commands involving piping from network tools, e.g., `curl | sh` or `wget | sh`, which are known bypass payloads.

Note that no patched version exists yet, so detection and monitoring are critical to identify exploitation attempts.

Useful 0 Not Useful 0

Compliance Impact

The provided information does not explicitly address the impact of CVE-2026-36045 on compliance with common standards and regulations such as GDPR or HIPAA.

Useful 0 Not Useful 0