ralph-wiggum

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.70). Most URLs are to reputable resources (official domains, Go and Rust installers, GitHub repos, DOI/openalex/press links), but the presence of direct installer scripts (https://sh.rustup.rs and https://claude.ai/install.sh) and direct archives that users may be instructed to curl|sh or run makes this set moderately risky if executed without verifying the contents and signatures.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill's Dockerfiles and build scripts run a remote installer (curl -fsSL https://claude.ai/install.sh | sh), which fetches and executes remote code to install the Claude CLI that the loop requires to run, so a required external script is executed during setup/build and enables runtime agent behavior.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly enables and documents use of a "--dangerously-skip-permissions" mode that can delete files, read sensitive credentials, run arbitrary commands, and even provides an override to run outside the sandbox—behavior that pushes the agent to bypass security and compromise host state.