Executive brief
Mosaxiv clawlet is an AI assistant tool that allows for the execution of system commands. A security flaw in its safety guard allows attackers to bypass command restrictions by wrapping dangerous commands inside other programs like Python. This could allow an attacker to delete files or execute unauthorized code on the system where the assistant is running, potentially leading to data loss or system compromise.
Technical details
A protection mechanism failure exists in the `guardExecCommand` function within `tools/tool_exec.go` (and `tools/exec_guard.go`) of mosaxiv clawlet. The safety guard uses regex-based denylisting to block dangerous shell commands like 'rm -rf', but it only evaluates the outer command string. An attacker can bypass these checks by wrapping malicious payloads inside inline interpreters (e.g., `python3 -c "..."`), which the guard does not normalize or unwrap. This allows for arbitrary command execution and filesystem modification within the workspace. The vulnerability is reachable via the exported `exec` tool interface, and while a GitHub issue was filed, the maintainers have marked it as 'not planned' for fixing.
Affected products
- mosaxiv clawlet <= 0.2.10
Timeline
- 2026-07-14: disclosed
- 2026-07-14: advisory: GitHub issue closed as 'not planned'