indirect-syscall-dev
Fail
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: HIGHCOMMAND_EXECUTIONOBFUSCATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill contains functional assembly code and implementation patterns for C, Rust, and Go to execute system calls indirectly. This technique is specifically designed to bypass user-mode hooks used by security software to monitor process behavior.
- [OBFUSCATION]: The instructions recommend and provide methods for hiding NT API names (like NtAllocateVirtualMemory) using seeded hashes and encrypted string tables. This is intended to evade automated analysis and human auditing of the resulting binary.
- [COMMAND_EXECUTION]: The skill implements logic to parse the internal structures of ntdll.dll at runtime to resolve system call numbers (SSNs). This allows an attacker to interact with the kernel directly, bypassing the standard Windows API layer.
- [INDIRECT_PROMPT_INJECTION]: The skill provides a significant attack surface by offering detailed templates for evasive loaders. An agent processing this skill lacks explicit constraints or boundary markers to prevent the generation of malicious software.
- Ingestion points: SKILL.md and multiple reference files containing implementation details.
- Capability inventory: Functional assembly trampolines, PE header parsing, and memory management API interaction.
- Boundary markers: Absent; no instructions provided to limit the use of these techniques to authorized research.
- Sanitization: Not applicable to static reference material.
Recommendations
- AI detected serious security threats
Audit Metadata