reverse-engineer-binary
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a rigorous process for analyzing untrusted binaries within hardened, air-gapped sandbox environments (e.g., gVisor, Firecracker). It explicitly prohibits network access during analysis and requires dropping Linux capabilities to ensure host security.\n- [COMMAND_EXECUTION]: The skill leverages command-line tools for disassembly, decompilation, and dynamic tracing (Ghidra, radare2, Frida). These operations are restricted to isolated guest environments to prevent potential host compromise from malicious artifacts.\n- [INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted binary data, it includes specific guardrails (ASI06) to prevent raw binary streams from entering the agent's prompt context, thereby mitigating the risk of instructions embedded within the analyzed data influencing agent behavior.
Audit Metadata