The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill executes remote code from the npm registry at runtime using the npx command.
Evidence: In scripts/shadcn_add.py, the command npx shadcn@latest add is constructed and executed. This pattern fetches the latest version of the tool from the internet every time a component is added, which could lead to unpredictable behavior if the remote package is updated or compromised.
COMMAND_EXECUTION (LOW): The skill utilizes subprocess.run to call external CLI tools.
Evidence: The script scripts/shadcn_add.py runs commands using subprocess.run with argument lists. While this is the recommended way to call system tools and prevents basic shell injection, it still provides the agent with a capability to execute processes on the host machine.
PROMPT_INJECTION (LOW): The skill provides an attack surface for indirect prompt injection because it ingests untrusted user input and passes it to system commands.
Ingestion points: User-provided component names in scripts/shadcn_add.py.
Boundary markers: None present in the logic to delimit user-controlled strings from instructions.
Capability inventory: The skill has the ability to execute system commands (npx) and interact with the local file system (components.json).
Sanitization: The script relies on the structural separation of arguments in subprocess.run; however, it does not validate or sanitize the content of the component name strings before execution.

ui-styling