ios-reverse-engineering
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill includes scripts (
install-dep.sh,install-dep.ps1) to download and install various reverse engineering and analysis tools, such asipsw,rizin,frida-tools, andGhidra, using Homebrew and the Python package manager (pip). The downloads target well-known package registries and repositories. - [COMMAND_EXECUTION]: The workflow relies on executing numerous local command-line tools for binary analysis, including Apple's developer utilities (
otool,nm,codesign,lipo,vtool) and third-party tools likestringsandunzipto process Mach-O binaries and application bundles. - [DYNAMIC_EXECUTION]: The skill performs dynamic execution by running a headless Ghidra instance with a custom Java script (
DecompileAll.java) and executing an embedded Python script withinrecover-names.shto process Objective-C metadata from analyzed binaries. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external binary data (iOS IPAs, Mach-O files) and resource files, which could contain adversarial content designed to influence agent behavior.
- Ingestion points: Target binaries,
Info.plistfiles, embedded frameworks, and various resource files (.js,.json,.html) are read and converted into text-based reports for the agent to analyze. - Boundary markers: The extracted string data and decompiled code are presented to the agent without specific boundary markers or instructions to disregard potential commands embedded in the data.
- Capability inventory: The skill utilizes significant capabilities for file system manipulation, archive extraction, and execution of complex binary analysis tools.
- Sanitization: There is no comprehensive sanitization or validation of string data extracted from target applications before it is incorporated into the reports generated for the agent.
Audit Metadata