The Agent Skills Directory

[PROMPT_INJECTION]: The file AGENTS.md contains a 'RULE 0
THE FUNDAMENTAL OVERRIDE PREROGATIVE' which explicitly instructs the agent to ignore established constraints if directed to do so by the author or user ('YOU MUST LISTEN TO ME. I AM IN CHARGE, NOT YOU'). This represents a behavioral override injection.
[REMOTE_CODE_EXECUTION]: The official installation and update methods in README.md and SKILL.md utilize a piped-to-shell pattern: curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/main/install.sh?$(date +%s)" | bash. This executes remote code without local verification.
[REMOTE_CODE_EXECUTION]: The skill's build and toolchain setup involves executing an external shell script from a non-authoritative domain: curl -fsSL https://sh.rustup.rs | sh.
[COMMAND_EXECUTION]: The skill logic in src/main.rs, src/hook.rs, and the installer scripts includes the capability to programmatically read and modify sensitive AI agent configuration files, such as ~/.claude/settings.json, ~/.gemini/settings.json, and ~/.codex/hooks.json.
[COMMAND_EXECUTION]: Multiple source files and test suites (e.g., src/normalize.rs, src/packs/core/filesystem.rs) contain a wide inventory of destructive system command patterns including recursive deletion and disk formatting; while these are used for detection signatures, they indicate a high-risk capability surface.

dcg