The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill utilizes npx create-dzx@latest to download and run the scaffolding tool. The source package create-dzx is not hosted by a recognized trusted organization, posing a risk of executing unverified third-party code.
COMMAND_EXECUTION (LOW): The instructions include multiple shell commands (dzx init, dzx validate, dzx inspect, dzx dev, dzx build) used for project lifecycle management.
REMOTE_CODE_EXECUTION (MEDIUM): Running npx with @latest downloads and executes code from a remote registry at runtime. This allows for the execution of potentially modified or malicious code if the registry or package is compromised.
PROMPT_INJECTION (LOW): The skill is susceptible to indirect prompt injection (Category 8) as it processes untrusted user input to define project architecture and manifest settings.
Ingestion points: User-provided purpose, workflows, and constraints in Step 1.
Boundary markers: Absent; there are no instructions to the agent to treat user input as data rather than instructions.
Capability inventory: File system writes (scaffolding), network access (npx), and arbitrary command execution (dzx dev/build).
Sanitization: Absent; the skill does not mention escaping or validating user inputs before interpolating them into the project structure or manifest.

create-project