The Agent Skills Directory

[DYNAMIC_EXECUTION]: The instructions direct the agent to write and execute 'discovery scripts' using the run_code tool at runtime to validate connectivity and explore data shapes.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates an attack surface where untrusted user requirements are processed to generate executable code. While the skill lacks explicit sanitization instructions for these inputs, it mandates a strict 'plan-and-confirm' phase requiring explicit human approval before any code is executed or changed.
Ingestion points: User integration requests and third-party API data fetched during discovery scripts (e.g., Shopify order data).
Boundary markers: The workflow uses 'confirm before changing' and 'explain before acting' as human-in-the-loop boundaries.
Capability inventory: Uses run_code for execution and yc_api_* tools for resource management.
Sanitization: Relies on manual user review of the plan and code before execution.
[COMMAND_EXECUTION]: The workflow leverages the run_code MCP tool to execute arbitrary JavaScript and Python code for validation and testing purposes.
[EXTERNAL_DOWNLOADS]: The example documentation references standard, well-known libraries such as @shopify/shopify-api and @sendgrid/mail as recommended dependencies for specific integration tasks.

fcode-agent