The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches real-time documentation as Markdown from the official vendor domain (neon.com). It also references the installation of the Neon MCP server from a vendor-controlled endpoint (mcp.neon.tech). These resources are used for up-to-date technical guidance.
[COMMAND_EXECUTION]: Provides commands to initialize projects and install extensions using npx and official CLI tools. These operations are intended for environment setup, including generating .env files and configuring MCP servers. The instructions explicitly advise handling credentials privately and using gitignore for sensitive files.
[INDIRECT_PROMPT_INJECTION]: The skill architecture involves fetching and processing external documentation content from live URLs, which represents a potential ingestion surface for untrusted data.
Ingestion points: Technical documentation fetched from neon.com/docs and index files like llms.txt.
Boundary markers: The skill relies on natural language instructions for the agent to verify claims against the official documentation but lacks formal data delimiters for ingested content.
Capability inventory: Capabilities include writing configuration files, performing network requests to fetch documentation, and executing local CLI commands for database management.
Sanitization: Instructions emphasize that documentation retrieval is for verification and that the agent should obtain user approval before performing destructive or costly actions.

neon-postgres