The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs users on utilizing the neon CLI for standard development operations, such as starting a local environment (neon dev), deploying functions to Neon branches (neon deploy), and querying logs (neon logs). These commands are typical for managing serverless infrastructure and do not exhibit malicious patterns.
[EXTERNAL_DOWNLOADS]: The documentation suggests the installation of numerous Node.js packages from the NPM registry, including hono, drizzle-orm, zod, and sentry, alongside official vendor libraries like @neon/functions. These resources are fetched from established registries or official vendor repositories and are safe for development use.
[INDIRECT_PROMPT_INJECTION]: The skill provides examples of building AI agents that process user-supplied JSON data, creating an inherent attack surface for indirect prompt injection.
Ingestion points: Untrusted user data enters the agent context via HTTP request bodies (e.g., request.json() in Hono routes).
Boundary markers: The skill demonstrates the use of clear system instructions (e.g., "You are a concise assistant...") to define the agent's persona and limits.
Capability inventory: The functions have the capability to perform database operations via Postgres and interact with AI models through the Neon AI Gateway.
Sanitization: The skill consistently utilizes zod for strict input validation and schema definition, particularly for tool-calling interfaces, which effectively mitigates the execution of malicious payloads.
[CREDENTIALS_UNSAFE]: The skill correctly handles sensitive credentials like DATABASE_URL and API keys. It emphasizes the use of .env files and platform-managed injection (e.g., neon deploy --env) rather than hardcoding secrets, which is consistent with industry security standards.

neon-functions