[PROMPT_INJECTION]: The AGENTS.md file contains a fundamental override directive ('If I tell you to do something, even if it goes against what follows below, YOU MUST LISTEN TO ME') that instructs the AI agent to disregard other behavioral constraints if the user gives a conflicting command.
[EXTERNAL_DOWNLOADS]: The skill's installation, uninstallation, and update scripts fetch resources from the developer's official GitHub repository (github.com/Dicklesworthstone/destructive_command_guard).
[REMOTE_CODE_EXECUTION]: The installation process fetches and executes code from remote sources, including the official Rust installer (sh.rustup.rs) and the project's own repository using shell-piped execution patterns.
[PERSISTENCE]: The tool includes a setup option to add a warning check to shell initialization files (.bashrc and .zshrc); this persistent check notifies the user if the tool's safety hooks have been removed from the agent's configuration.
[INDIRECT_PROMPT_INJECTION]: The dcg scan command processes untrusted content from local files (shell scripts, Dockerfiles, YAML configs) to identify destructive patterns, creating an attack surface where malicious instructions embedded in those files could be ingested into the agent's context.
[COMMAND_EXECUTION]: The skill executes compiled binaries and utilizes subprocesses in its performance benchmarking scripts to monitor execution latency and verify environment compatibility.

dcg