The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is designed to process and act upon untrusted external data including C++ source files and CMake build configurations.
Ingestion points: File triggers for .cpp, .hpp, CMakeLists.txt, vcpkg.json, and conanfile.txt.
Boundary markers: None identified in the prompt templates or instructions to distinguish between developer intent and embedded malicious data.
Capability inventory: The skill is explicitly allowed to use Bash with g++, gcc, clang, cmake, make, ctest, and valgrind. These tools can execute arbitrary code during the build or test phases.
Sanitization: No evidence of sanitization for compiler flags or CMake commands derived from input data.
[Unverifiable Dependencies & Remote Code Execution] (MEDIUM): The skill promotes the use of external package managers and build-time downloads.
Evidence: Recommends using FetchContent in CMake to download and execute code from remote repositories and vcpkg/Conan for package management.
Risk: Malicious repositories could provide poisoned dependencies that execute code during the build process.
[Command Execution] (LOW): The skill requires extensive use of shell commands for compilation and debugging. While this is the intended purpose, it provides a powerful primitive for an attacker who successfully injects instructions.

moai-lang-cpp