The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Recommends installing the @modelcontextprotocol/sdk and zod libraries from the official npm registry for Node.js implementations.
[EXTERNAL_DOWNLOADS]: Suggests downloading and installing the mcptools package from the Posit (formerly RStudio) official GitHub repository for R-based servers.
[COMMAND_EXECUTION]: Instructs the user to use the claude mcp add command to register the newly created server with the local environment, which involves executing shell commands to link the server script.
[DATA_EXFILTRATION]: Provides examples of tools designed to query databases and return structured data to the AI assistant. The skill includes a 'Pitfalls' section that explicitly warns users to carefully validate inputs for database queries and shell commands to prevent unauthorized access.
[PROMPT_INJECTION]: Contains a vulnerability surface for indirect prompt injection via the query_database and run_analysis tools.
Ingestion points: Tool parameters query, analysis_name, and dataset in SKILL.md.
Boundary markers: The Node.js example implements a regex check (/^\s*SELECT/i) to ensure only read-only queries are accepted.
Capability inventory: The tools utilize DBI::dbGetQuery (R) or executeQuery (Node.js) to access external data sources.
Sanitization: While the Node.js example uses regex validation, the R example uses simple string concatenation for the query construction.

build-custom-mcp-server