The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a method for generating AI assistant tools based on external OpenAPI specifications. This process creates a surface for indirect prompt injection, as descriptions or metadata within a malicious specification could be used to manipulate the behavior of an agent utilizing the generated tools.
Ingestion points: OpenAPI specification files or URLs defined as inputs in the .speakeasy/workflow.yaml configuration.
Boundary markers: The workflow instructions suggest running speakeasy lint to validate the specification, which helps ensure schema validity but does not verify the safety of natural language content.
Capability inventory: The resulting MCP server enables tools that allow an agent to execute network operations (API calls) as defined in the target specification.
Sanitization: The instructions do not detail any explicit sanitization or filtering of the text content from the specification during the generation of tool prompts.
[COMMAND_EXECUTION]: The skill workflow involves executing commands via the speakeasy CLI and the npx package runner. These commands are part of the standard, expected operations for using the vendor's toolset and are documented for legitimate development use.

generate-mcp-server