The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides patterns for handling untrusted data that represent an indirect prompt injection surface.\n
Ingestion points: Untrusted data enters the application via formData in references/server-actions.md and through dynamic route parameters (params) in references/app-router.md.\n
Boundary markers: No specific boundary markers or instructions to ignore embedded prompts are provided in the code templates.\n
Capability inventory: The provided examples include file system writes (fs/promises.writeFile in references/server-actions.md) and database mutations (prisma.db.post.create in references/server-actions.md).\n
Sanitization: While some examples utilize zod for basic type validation, others lack sanitization, specifically for filenames in file operations.\n- [COMMAND_EXECUTION]: A code example in references/server-actions.md demonstrates an insecure file upload implementation.\n
Evidence: The uploadAvatar function uses file.name directly within path.join(process.cwd(), 'public', 'uploads', file.name). This allows for a path traversal attack where an attacker could overwrite sensitive files outside the intended directory by providing a manipulated filename (e.g., ../../target-file).\n- [EXTERNAL_DOWNLOADS]: The skill references and installs dependencies from well-known technology services.\n
Evidence: Mentions of npm install -g vercel, @sentry/nextjs, and @vercel/analytics are standard for Next.js development and originate from established vendors.\n- [SAFE]: The core workflow and architectural constraints defined in SKILL.md align with current industry best practices for Next.js 14, including the use of Server Components by default and proper environment variable management.

nextjs-developer