The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via its site-scraping feature. It retrieves content from user-provided URLs using curl and subsequently analyzes this content to generate a project brief. Malicious instructions embedded in the scraped HTML could be processed by the agent, potentially leading to unauthorized operations.
Ingestion points: SKILL.md Fase 0.A.1 scraping logic.
Boundary markers: None present to isolate scraped content from agent instructions.
Capability inventory: High, including shell access (npx, npm), filesystem operations, and network requests.
Sanitization: No sanitization of the scraped data is described before analysis.
[EXTERNAL_DOWNLOADS]: The skill downloads code components from external registries during project construction.
Evidence: references/component-catalog.md lists magicui.design, reactbits.dev, 21st.dev, and ui.aceternity.com as sources for npx shadcn@latest add commands.
Context: These are well-known services within the web development ecosystem, but fetching code directly from remote URLs bypassing standard package verification is a notable behavior. Community-authored content on platforms like 21st.dev carries higher inherent risk.
[COMMAND_EXECUTION]: The skill makes extensive use of terminal commands to automate the development workflow.
Evidence: SKILL.md contains commands for scraping (curl), project initialization (npx create-next-app), and dependency management (npm install).
Context: These commands are standard for the tool's stated purpose but provide a mechanism for potential exploitation if the agent is compromised via indirect prompt injection.

frontend-design-engineer