The Agent Skills Directory

Dynamic Execution of Generated Content: The skill generates JavaScript implementation files based on user-supplied requirements and subsequently validates the generated code by importing it as an ESM data URL and executing the runTool function using Node.js. While this is performed within a mock environment to verify functionality, executing dynamically generated content is a pattern that requires careful management of the execution context.
Command Execution via External Tooling: The skill utilizes the Power Platform CLI (pac) to perform Dataverse table discovery and type generation (pac model list-tables, pac model genpage generate-types). This involves executing system-level commands to interact with environment data and generate temporary type definitions.
Indirect Prompt Injection Surface: As the skill processes user-supplied input (such as tool purpose, input fields, and expected results) to construct executable JavaScript logic, there is a potential for untrusted input to influence the structure or behavior of the generated artifacts. The skill includes safeguards by instructing the agent to validate all toolInput and escape values interpolated into OData filters within the generated code.
Ingestion points: User-supplied tool purpose, input field definitions, and expected result descriptions (SKILL.md).
Boundary markers: The skill uses structural boundaries by requiring specific basenames for the matched pair of files (.tool.js and .tool.json) and enforcing valid JSON schemas for input/output contracts.
Capability inventory: File system operations (Read, Write, Edit), command-line execution via Bash and the pac CLI, and dynamic JavaScript execution via Node.js (SKILL.md).
Sanitization: The instructions explicitly require the generated code to validate all toolInput and escape OData filter values to prevent injection within the generated tool's runtime context (SKILL.md).

generate-codeful-mcp-tool