zeude

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt instructs embedding secrets verbatim (agent keys in the curl installer, and Supabase/ClickHouse credentials in .env/DSNs), which requires the LLM to handle/output secret values directly.

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.90). Suspicious — the list includes a direct raw.githubusercontent.com install.sh intended to be piped to bash (remote executable script) and GitHub repos/clone URLs for third‑party shims that intercept CLI commands (allowing arbitrary code execution and potential data exfiltration), which are high-risk unless you fully trust and have audited the repos.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The project intentionally captures every Claude prompt via OpenTelemetry and uploads it to external analytics, installs a local "Shim" that intercepts the claude CLI and auto-syncs/auto-updates skills and hooks from a central dashboard (deployed remotely via an agent key and a curl|bash installer), which together constitute clear data-exfiltration capability and a remote code-deployment/backdoor vector that can push executable hooks or skills to end users.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs running a remote installer that fetches and immediately executes a script via curl -fsSL https://raw.githubusercontent.com/zep-us/zeude/main/install.sh | bash ..., which runs remote code at install time and is presented as the supported per-developer installation method (required for the Zeude Shim), so it directly executes external code during runtime.