The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the zapier-sdk command-line interface to perform administrative tasks such as listing apps, managing connections, and discovering app actions.
[REMOTE_CODE_EXECUTION]: The skill generates TypeScript code (workflow.ts) and executes it within a sandbox environment using the zapier-sdk --experimental run-durable command. This is the primary function of the skill for testing workflows.
[EXTERNAL_DOWNLOADS]: A Node.js script is used to query the npm registry via npm view to identify specific package versions that meet age requirements for the execution sandbox. It also pins dependencies for @zapier/zapier-sdk, @zapier/zapier-durable, and zod.
[PROMPT_INJECTION]: The skill processes natural language requests to define workflow logic, which constitutes a surface for indirect prompt injection.
Ingestion points: Phase 1 (Natural language request ingestion in SKILL.md).
Boundary markers: None present to delimit user instructions from system logic.
Capability inventory: File system access (writing workflow.ts, package.json), network access (via zapier-sdk and connected apps), and command execution.
Sanitization: No specific sanitization or validation of the natural language input is performed before it is used to generate workflow code, though human review is required in Phase 3.

workflows-create