The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates (such as assets/openapi-integration.py and documentation in references/integration-patterns.md) that allow an agent to auto-generate MCP tools and resources by ingesting external OpenAPI specifications.
Ingestion points: The template in assets/openapi-integration.py uses httpx.get(OPENAPI_SPEC_URL) to fetch external specifications at runtime.
Boundary markers: The code that processes these specifications and defines tool schemas does not include boundary markers or explicit instructions to the LLM to ignore potentially malicious commands embedded in the documentation fields (descriptions, summaries) of the ingested specification.
Capability inventory: The generated server templates possess the capability to perform network operations using httpx.AsyncClient and potentially interact with the file system or sensitive data depending on the tools defined in the external specification.
Sanitization: There is no evidence of sanitization, filtering, or validation of the natural language content within the fetched OpenAPI specifications before they are interpolated into the agent's context and tool definitions.

mcp-builder