The Agent Skills Directory

[Unverifiable Dependencies & Remote Code Execution] (CRITICAL): The skill documentation (specifically examples/install-update-script.md) directs the agent to perform piped remote execution, commonly curl ... | bash. This pattern allows an untrusted remote source to execute arbitrary code with the user's current shell permissions.
[Persistence Mechanisms] (HIGH): The skill manages shell integration (examples/shell-integration.md) and persists custom settings in shell profiles (examples/colors-persist.md). This involves modifying files like ~/.bashrc, ~/.zshrc, and ~/.profile, which are primary vectors for maintaining persistent access to a system.
[Indirect Prompt Injection] (HIGH): The skill processes external .nvmrc files (examples/nvmrc.md).
Ingestion points: Reading .nvmrc files from user-controlled or project-specific directories.
Boundary markers: None identified in the skill metadata.
Capability inventory: Executing shell commands, switching Node.js versions, and managing global packages.
Sanitization: No evidence of sanitization for the contents of .nvmrc before processing.
[Data Exposure & Exfiltration] (MEDIUM): The skill handles authentication headers for custom mirrors (examples/mirror-auth-header.md). If an attacker influences the mirror URL via an environment variable or configuration file, sensitive credentials/tokens could be exfiltrated to a malicious server.
[Privilege Escalation] (MEDIUM): While nvm generally installs to a user's home directory, the troubleshooting and Alpine install guides (examples/alpine-install.md, examples/troubleshooting-linux.md) often involve commands that may require elevated privileges or modify system-level paths.

nvm