reverse-engineering-advanced
Fail
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the user and agent to execute commands with elevated privileges using
sudofor tool installation and environment setup. Specific instances includesudo apt install,sudo dpkg -i,sudo ./installer.sh, andsudo make installfound inguides/reverse-engineering-advanced-playbook.mdandSKILL.md. - [EXTERNAL_DOWNLOADS]: The skill fetches code and tools from various untrusted third-party GitHub repositories and external sources that are not on the trusted vendor list. Examples include
git clone https://github.com/cd70s062f/deflat,git clone https://github.com/McGill-DMaS/Kam1n0-Community,git clone https://github.com/fkie-cad/FACT_core, andgit clone https://github.com/e-m-b-a/emba. While it also uses trusted sources like Google/Zynamics for BinDiff, the reliance on unknown community repositories increases the supply chain risk. - [COMMAND_EXECUTION]: The skill frequently executes system commands and spawns subprocesses to analyze user-provided binary files. Tools like
nm,readelf,objdump,binwalk, andr2are invoked with arbitrary filenames, which could be exploited if filenames are not properly sanitized or if the tools themselves contain vulnerabilities. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted binary data from external sources. It lacks boundary markers or explicit instructions to ignore malicious data embedded in these binaries. Given its capabilities (file writing, network access for updates, and command execution), a malicious binary could potentially influence agent behavior or exploit known vulnerabilities in the underlying RE tools (e.g., Ghidra's CVE-2024-24582 mentioned in the skill's own case studies).
- [DYNAMIC_EXECUTION]: The skill generates and executes Python scripts at runtime to perform symbolic execution and binary analysis using the
angrandz3libraries. It also includes instructions for writing and executing Ghidra and IDA Python scripts, which involves dynamic assembly and execution of code based on analyzed binary content.
Recommendations
- AI detected serious security threats
Audit Metadata