relay-to-agent_skill

Overview

This skill relays messages to AI agents on any OpenAI-compatible API and manages multi-turn conversations with persistent sessions. It lists available agents, sends messages, and can reset or isolate sessions for parallel dialogues. Built to work with many endpoints like OpenRouter, LiteLLM, vLLM, Ollama, and other Chat Completions implementations. Configuration is file- and environment-driven for flexible deployment.

How this skill works

The tool reads agent definitions from a local JSON config and sends chat messages to the configured base URL using an API key. Sessions are tracked locally so agents can remember conversation context across multiple invocations. You can reset a session to clear history, specify a custom session ID for concurrent conversations, or request raw JSON output for integrations. It supports listing agents, sending one-off messages, and continuing multi-turn exchanges.

When to use it

• When you need to forward user prompts to a named agent on any OpenAI-compatible endpoint.
• To maintain multi-turn context for follow-up questions or conversational workflows.
• When running agents locally (vLLM, LiteLLM, Ollama) or via hosted routers (OpenRouter, Connect Chat).
• When you require simple session isolation for parallel conversations with the same agent.
• When integrating agent messaging into scripts, CI jobs, or automation pipelines.

Best practices

• Store API keys in environment variables and avoid committing credentials to config files.
• Keep agents.json under version control for reproducible agent definitions, but exclude secrets.
• Use custom session IDs for separate threads of conversation to prevent context leakage.
• Limit message history size or reset sessions regularly to control token usage and memory.
• Enable raw JSON output for programmatic parsing and logging in automation workflows.

Example use cases

• Transform an article into a LinkedIn post by relaying the content to a marketing agent.
• Analyze campaign performance and then ask follow-ups that reference previous responses.
• Run local model experiments with vLLM or Ollama while keeping conversations persistent.
• List all deployed agents and their descriptions to choose the right model for a task.
• Reset a session to start a fresh conversation when switching topics or workflows.