create-agent_skill

Overview

This skill bootstraps a modular AI agent using the OpenRouter SDK with an extensible hook system and an optional Ink-based terminal UI. It provides a standalone agent core that can run headless, register tools at runtime, and stream items-based responses for robust UI updates. Use it to quickly integrate 300+ models, tool calls, and progressive reasoning into your application.

How this skill works

The agent core wraps the OpenRouter client and exposes a lifecycle and event hooks for messages, streaming items, tool calls, reasoning updates, and errors. Streaming uses an items-based model: UI layers replace items by ID instead of concatenating chunks, enabling accurate partial updates. Tools are defined with zod schemas and can be added at runtime; optional Ink TUI consumes the agent events to render a responsive terminal chat.

When to use it

• Building a headless chatbot or bot service that needs streaming and tool integration.
• Embedding an AI assistant into a terminal app with a polished Ink UI.
• Creating orchestration logic that needs hooks for tool calls, reasoning, and partial outputs.
• Prototyping multi-model workflows using OpenRouter’s unified model access.
• Systems that require replace-by-ID streaming updates for reliable UI rendering.

Best practices

• Keep API keys in environment variables and never commit them to source control.
• Use items-based streaming for UIs to replace items by ID rather than accumulating raw chunks.
• Define tools with zod schemas to validate inputs and outputs and reduce runtime errors.
• Limit step count / maxSteps to avoid runaway tool loops and long reasoning chains.
• Hook into error, thinking, and stream events for clear UX and telemetry.

Example use cases

• A CLI knowledge assistant that calls web search and calculator tools and shows progressive answers in the terminal.
• A Discord bot that routes user inputs to the agent core and streams responses back to chat with incremental updates.
• A backend service that runs headless agent workflows and triggers external APIs via tool calls.
• Rapid prototyping of model behaviors by swapping models through OpenRouter without changing agent logic.