cpp-cleanup

Fail

Audited by Gen Agent Trust Hub on May 10, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Instructions in SKILL.md (Step 1) direct the agent to compile C++ source files using g++ and then execute the resulting binary (./a.out). Running binaries compiled from untrusted source code is a high-risk operation that enables arbitrary code execution within the agent's environment.
  • [COMMAND_EXECUTION]: The skill invokes several shell-based tools (clang-tidy, cppcheck, g++) to process local files. Executing shell commands on untrusted directories poses a risk of command injection or exploitation of the tools themselves.
  • [PROMPT_INJECTION]: The skill ingests untrusted C++ source code, creating a surface for indirect prompt injection.
  • Ingestion points: C++ source files (*.cpp) targeted by the tools in Step 1.
  • Boundary markers: None identified; the skill does not instruct the agent to treat comments or strings within the code as untrusted.
  • Capability inventory: The skill combines data ingestion with compilation (g++) and direct binary execution (./a.out), establishing a path from data input to machine code execution.
  • Sanitization: No validation or scanning of the source code is performed to detect malicious logic or embedded instructions before compilation or execution.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
May 10, 2026, 05:22 PM
Security Audit — agent-trust-hub — cpp-cleanup