dynamic-linking

Fail

Audited by Gen Agent Trust Hub on Feb 21, 2026

Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [Privilege Escalation] (HIGH): The skill explicitly directs the agent to execute 'sudo cp' and 'sudo ldconfig' (SKILL.md, references/ld-rpath-soname.md) to modify system-wide library paths and caches, which presents a significant privilege escalation risk.
  • [Dynamic Execution] (MEDIUM): The skill provides comprehensive guides for 'dlopen' and 'LD_PRELOAD' interposition (SKILL.md), which are powerful mechanisms for runtime code loading and function interception that can be exploited for malicious behavior.
  • [Indirect Prompt Injection] (LOW): The skill provides instructions to process potentially untrusted ELF binaries using 'readelf', 'nm', and 'objdump' without input sanitization or boundary markers. 1. Ingestion points: Local binary files provided for analysis. 2. Boundary markers: Absent. 3. Capability inventory: System-level command execution (sudo), C compilation (gcc), and dynamic loading. 4. Sanitization: Absent.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 21, 2026, 11:05 AM