The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md references www.shadcn-vue.com as an official resource. This domain is currently flagged by security scanners as a malicious 'CryptScam' site. Directing users to malicious domains for project documentation or resources is a severe security concern.
[COMMAND_EXECUTION]: Scaffolding scripts such as scripts/shadcn_vue_init.py and scripts/electron_vue_init.py are vulnerable to command injection on Windows. They use subprocess.run(..., shell=True) with a project-name argument that is derived from user input or context without sanitization. An attacker could provide a project name containing shell escape sequences (e.g., & calc.exe) to execute arbitrary code on the host machine.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external URLs to synchronize design tokens. This represents a vulnerability surface for indirect prompt injection.
Ingestion points: SKILL.md (Step 2) fetches a DSL JSON from a user-specified URL or environment variable.
Boundary markers: None identified; the instructions do not implement markers to separate instructions from data.
Capability inventory: The skill uses the get_token tool to inject DSL data into Tailwind CSS configurations and subsequently executes the development server via bun run dev.
Sanitization: None; the skill assumes the remote DSL is safe and well-formatted without validation.
[EXTERNAL_DOWNLOADS]: The scripts hardcode external mirrors for npm registries and Electron binaries (e.g., registry.npmmirror.com). While these are common mirrors, they represent a third-party supply chain dependency outside of official package registries.
[COMMAND_EXECUTION]: The workflow involves automated, high-privilege operations including running third-party CLI tools via npx (npx shadcn-vue@latest) and executing installation scripts. Without strict validation of the inputs and sources, this automation facilitates the execution of potentially malicious code from external sources.

vue-creater