The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a provided diagnostic script scripts/verify-setup.sh. This script is used to verify the local environment (Tailwind configuration, components.json, and path aliases) and is restricted to read-only operations on project metadata. Additionally, the skill utilizes standard CLI tools including npx, npm, and git for project initialization and component management.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection as it is designed to ingest and process Stitch HTML and screenshots (as noted in SKILL.md and references/workflow.md).
Ingestion points: External HTML and screenshot assets provided by the user or fetched via MCP.
Boundary markers: Explicit instructions are provided to the agent to verify inputs and treat assets within the context of the requested scope only.
Capability inventory: The agent can perform file writes, execute a local bash script, and run npm/npx commands.
Sanitization: The skill includes mandatory instructions in SKILL.md and references/faq-deep.md to remove personally identifiable information (PII), cookies, and sensitive tokens from processed assets, replacing them with demonstration values.
[EXTERNAL_DOWNLOADS]: The skill frequently references npx shadcn@latest and npm install for fetching components and managing dependencies. These operations target well-known, official registries and repositories for the shadcn/ui framework, which are recognized as safe sources for development utilities.

stitch-shadcn-ui