refly

The Refly base skill serves as the core orchestration layer for the Refly ecosystem. It is designed for developers, automation engineers, and power users who need to transform complex business logic and Standard Operating Procedures (SOPs) into deterministic, version-controlled agent skills. By acting as the central interface, it allows users to manage their skill registry, discover new capabilities, and invoke workflows seamlessly through a unified command-line experience. It bridges the gap between raw API calls and high-level agentic intent, ensuring that automated tasks remain reliable, auditable, and easily maintainable within any development environment.

• Orchestrates skill discovery, installation, and execution across domain-specific workflow directories.

• Implements a robust stateful runtime that supports file generation, text processing, and multi-step action execution.

• Provides deterministic workflow management with built-in status monitoring, detail retrieval, and toolcall output handling.

• Supports seamless integration with external frameworks like Claude Code, MCP servers, and custom API backends.

• Features an intervenable runtime allowing for auditability and manual re-steering of long-running automated tasks.

• Always utilize the refly CLI commands (e.g., refly skill run, refly workflow status) rather than interacting with private APIs directly.

• Ensure the @refly/cli package is installed and properly authenticated before initiating workflows.

• Follow the defined execution patterns for specific output types: Pattern A (File Generation/Downloads), Pattern B (Text/Data Extraction), or Pattern C (Action/Integration Status).

• Use the --watch and --interval flags to monitor workflow progress effectively, and handle IDs (such as skpi-xxx for skills and we-xxx for workflow executions) with precision.

• Always stop execution and inspect the hint provided if a command returns ok=false; avoid hard-coding IDs or fabricating metadata to maintain infrastructure stability.