The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill takes user input from $ARGUMENTS and uses it to name a directory in a shell command (npm create vite@latest <name>). While the instructions direct the agent to strip special characters, there is a risk of command injection if the agent fails to sanitize the input properly before execution.
Ingestion points: Project name/description from $ARGUMENTS (SKILL.md).
Boundary markers: None implemented at the system level; instructions rely on agent adherence to sanitization rules.
Capability inventory: The skill uses Bash to create directories, install packages, and initialize the project environment.
Sanitization: Prompt-based instruction to "Strip spaces and special characters," which is a weak enforcement mechanism.
[COMMAND_EXECUTION]: The skill executes several bash commands to set up the development environment, including npm install and npx commands. These are legitimate for the skill's purpose but require user oversight.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs several packages from the npm registry, including well-known tools like vite and official vendor packages under the @evenrealities namespace. These resources are consistent with the vendor's own ecosystem.

quickstart