kernel-exploitation
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill includes instructions to download the Linux kernel source code directly from official Kernel.org CDN mirrors.
- [PRIVILEGE_ESCALATION]: Provides multiple methods for gaining root access via kernel vulnerabilities, including a modprobe_path overwrite demonstration in SKILL.md that uses chmod 777 to alter target file permissions.
- [COMMAND_EXECUTION]: Extensive instructions for executing shell commands to compile the kernel, orchestrate QEMU emulation, and perform low-level debugging with GDB.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from kernel-mode outputs and system interfaces which could potentially be used to influence the agent's behavior. * Ingestion points: Reading output from /proc/kallsyms, dmesg logs, and physical memory in SKILL.md and KERNEL_MITIGATION_BYPASS.md. * Boundary markers: No delimiters or warnings are present to isolate the agent from untrusted kernel output. * Capability inventory: Includes the ability to execute shell commands, perform ROP-based code execution, and download external files. * Sanitization: No explicit validation or sanitization of ingested kernel telemetry is described.
Audit Metadata