The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the Zapier platform using the zapier-sdk CLI to perform discovery, connection management, and workflow deployment.\n- [DYNAMIC_EXECUTION]: A Node.js script is executed to check the npm registry and select versions that are at least 24 hours old to ensure compatibility with the sandbox environment.\n- [EXTERNAL_DOWNLOADS]: The skill manages dependencies from the npm registry, including official @zapier packages and zod. These are either vendor-owned or well-known public packages.\n- [REMOTE_CODE_EXECUTION]: Generated workflow code is executed using the experimental run-durable command of the Zapier SDK, which facilitates testing in a controlled sandbox.\n- [INDIRECT_PROMPT_INJECTION]: The skill creates executable code based on natural language user requests, which is a potential injection surface.\n
Ingestion points: The user's natural language request defines the workflow logic in Phase 1.\n
Boundary markers: The agent is instructed to summarize the plan and obtain explicit user confirmation before generating any code.\n
Capability inventory: The skill can write files to the filesystem and execute SDK commands that interact with external services via the user's account.\n
Sanitization: No explicit sanitization of the natural language input is described in the instructions.

workflows-create