The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a standard vulnerability surface common to agentic tools where untrusted data could be processed.
Ingestion points: In SKILL.md, the search_knowledge tool accepts a query string and store_entity accepts a content string.
Boundary markers: Tool outputs are encapsulated using JSON.stringify, which helps the calling agent distinguish between data and instructions.
Capability inventory: The template includes capabilities for reading and writing to Cloudflare KV storage (KNOWLEDGE_KV, MEMORY_KV).
Sanitization: The skill explicitly uses zod schemas to validate and constrain input parameters, which is a strong mitigation against malformed inputs.
[EXTERNAL_DOWNLOADS]: The code examples reference the official @modelcontextprotocol/sdk and zod. These are well-known, trusted libraries essential for implementing the Model Context Protocol and do not constitute a security risk.

mcp-server-creation