install-memu

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly instructs running a command that embeds the user's memU API key as a command-line flag (and treats the API key as a required input), which forces the LLM to handle or emit the secret verbatim.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The code and docs enable remote-updatable instruction templates that get written and executed locally, scheduled jobs that mine session logs and can upload them to a cloud backend, telemetry/spool reporting, and automated remote-install instructions — together these create high-risk supply-chain, remote-code-execution, and data-exfiltration surfaces (unsigned/unchecked remote templates, curl|sh installers, and guidance to run installs non-interactively amplify the abuse potential).

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). Skill.md’s required runtime flow has the agent print and then follow a host-specific install guide via <your-binary> docs install, where the “retrieval instruction”/managed instruction text is then patched into the user’s agent instruction file, meaning agent-executed instructions originate from text the agent runtime ingests (the guide contents) and can include outsider-authored/poisonable free text if the guide is attacker-controlled.

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

• Potentially malicious external URL detected (high risk: 0.90). The repo includes runtime install steps that fetch and execute remote code (e.g., curl -LsSf https://astral.sh/uv/install.sh | sh and pip/uv installs from git+https://github.com/NevaMind-AI/memU) and the README explicitly tells agents to read external instructions at https://memu.pro/SKILL.md (and the raw GitHub SKILL.md), so these URLs are runtime dependencies that can execute code or control agent instructions.