binary-reverse
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The file
guides/Binary_Analysis_Reverse_Engineering_Story.mdcontains a shell command that pipes a hardcoded password string intosudoas part of a tutorial narrative. - Evidence:
sudo -S apt install -y ghidra <<< "security-platform.cn"inguides/Binary_Analysis_Reverse_Engineering_Story.md. - [INDIRECT_PROMPT_INJECTION]: The skill is primarily designed to ingest and analyze untrusted binary files. These files are external data sources that could contain malicious strings or code comments designed to mislead the AI agent during disassembly and analysis.
- Ingestion points:
Readtool used on binary files, output fromstrings, andradare2analysis data inSKILL.mdandpayloads.md. - Boundary markers: Absent. There are no instructions to the agent to disregard potential natural language commands found within the binary data.
- Capability inventory: The skill utilizes powerful tools including
Bash,WebFetch, and file systemWrite/Editcapabilities across several scripts. - Sanitization: No explicit sanitization or filtering is implemented for data extracted from binary symbols or strings before it is presented to the LLM.
- [COMMAND_EXECUTION]: The automation scripts provided in
payloads.mduse Python'ssubprocessandos.systemmodules to execute shell commands with variables derived from the binaries being analyzed. - Evidence: The
scan_binaryandfind_gadgetsscripts inpayloads.mdexecute shell-based analysis routines viar2pipe. - [DYNAMIC_EXECUTION]: The skill makes extensive use of symbolic execution and dynamic instrumentation frameworks (angr, Frida, Manticore) to analyze and potentially execute code branches within target binaries based on runtime state.
- Evidence: Scripts in
payloads.mdandguides/anti-reversing-bypass-guide.mddemonstrate the use ofangrfor constraint solving andFridafor memory monitoring and hooking.
Audit Metadata