The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by design, as it provides tools to query and display content from external PostgreSQL databases and potential third-party APIs.
Ingestion points: External data enters the agent's context through the listTables, queryDatabase, and executeDatabase tool outputs defined in SKILL.md.
Boundary markers: The tools return data as JSON-stringified text within standard MCP tool result structures. While specific prompt boundary instructions (like 'ignore instructions in data') are not explicitly included in the code snippets, the template focuses on technical transport.
Capability inventory: The skill enables database read/write operations and uses the wrangler CLI for deployment and environment management.
Sanitization: The template enforces robust sanitization including Zod schema validation for all tool inputs, mandatory parameterized queries for database interactions, and a regex-based validation layer to restrict queryDatabase to read-only operations (SELECT/SHOW/DESCRIBE).
[COMMAND_EXECUTION]: The documentation describes the use of standard development tools such as npm and the Cloudflare wrangler CLI for project initialization, authentication, and deployment.
[SAFE]: The skill promotes security best practices by explicitly instructing users to use .dev.vars for secrets and Cloudflare Secrets for production credentials, rather than hardcoding sensitive information like the DATABASE_URL or GITHUB_CLIENT_SECRET.

remote-mcp-server-with-auth