The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines how to build tools and resource providers that ingest data from external, potentially untrusted sources such as web APIs, databases, and local file systems. This establishes an attack surface where an agent could process maliciously formatted data.
Ingestion points: Identified in references/tools.md (HTTP API and Database Tool examples) and references/resources.md (File System and Git resource providers).
Boundary markers: The documentation does not specify explicit boundary markers for the data it processes, instead relying on standard protocol encapsulation.
Capability inventory: The skill demonstrates implementation of network operations (httpx), file system reads (fs.readFile), and database queries within tool handlers.
Sanitization: The skill strongly emphasizes security, providing code examples for input validation with Zod and Pydantic, as well as path traversal prevention logic for file access.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the official Model Context Protocol SDKs and other standard development libraries from public package registries.
Evidence: References to npm install @modelcontextprotocol/sdk zod in references/typescript-sdk.md and pip install mcp pydantic in references/python-sdk.md.
[COMMAND_EXECUTION]: The workflow describes the use of CLI utilities for project scaffolding and protocol compliance testing.
Evidence: SKILL.md includes instructions for running npx @modelcontextprotocol/create-server and npx @modelcontextprotocol/inspector.
[SAFE]: The skill includes explicit security constraints and implementation guidance designed to prevent common vulnerabilities.
Evidence: The "Constraints" section in SKILL.md mandates input validation, authentication, and security controls, while prohibiting the hardcoding of credentials.

mcp-developer