agent-designerlisted

Design an AI agent or bot before it meets users. Use it at design time, when the agent's behaviour is still on paper. It runs three methods from the UX + AI connector: one defines what the agent may do, must ask about first, and never does; one decides when it hands off to a human and what context travels with the handoff; one delivers its personality, voice, and behaviour as a usable system prompt. Outcome: an enforceable scope, a handoff plan, and a system prompt that holds the voice. Trigger on 'we are building an agent', 'our bot needs a personality', 'what should it sound like', 'it should do this automatically', 'how much should it be allowed to do', 'when should it ask permission', 'when does a human take over', 'give it a voice', or any request to scope, voice, or design the behaviour of something that acts on its own.
IleanaMarcut/uxai-mcp-plugin · ★ 1 · AI & Automation · score 64

Install: claude install-skill IleanaMarcut/uxai-mcp-plugin

# Agent Designer ## What this is Three methods for designing an AI agent or bot: the scope of what it may do, the handoff to a human when it cannot finish, and its personality delivered as a usable system prompt. The outcome is an agent whose behaviour is decided on paper before it acts in front of anyone. ## When and why to use it Use it at design time, while the agent's behaviour can still change cheaply. An agent acts while nobody is watching, so what it is allowed to do, when it hands off, and how it sounds have to be decisions, and this order makes each one before the next needs it. ## This skill runs the methods of the UX + AI MCP Each step is a method served by the UX + AI connector. This file only routes; the methods do the work, and they improve on the connector without this file changing. ## How it works 1. **Check the connector.** The UX + AI MCP must be connected on a subscribed account. If it is unavailable, say so and stop. 2. **Establish what the agent does.** Keep the first reply short: ask what it acts on and for whom, then stop. 3. **Fetch by id and follow the text exactly as returned.** These are the three methods this skill runs; never rebuild one from memory or from this file. ``` fetch(id: "agentic-scope") fetch(id: "escalation-triggers") fetch(id: "ai-agent-personality") ``` 4. **Run the steps in this order:** - **Scope** (`agentic-scope`): what the agent can do, must ask about first, and never does. Every later decision needs