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.md contains a shell command that pipes a hardcoded password string into sudo as part of a tutorial narrative.
  • Evidence: sudo -S apt install -y ghidra <<< "security-platform.cn" in guides/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: Read tool used on binary files, output from strings, and radare2 analysis data in SKILL.md and payloads.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 system Write/Edit capabilities 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.md use Python's subprocess and os.system modules to execute shell commands with variables derived from the binaries being analyzed.
  • Evidence: The scan_binary and find_gadgets scripts in payloads.md execute shell-based analysis routines via r2pipe.
  • [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.md and guides/anti-reversing-bypass-guide.md demonstrate the use of angr for constraint solving and Frida for memory monitoring and hooking.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM