Claude Agent Builder

Use when you want to design a repeatable agent workflow (e.g. “review PR”, “summarize docs”) for Claude or Claude Code.

Why useful

Agents need clear triggers, steps, and outputs. This prompt turns a vague goal into a concrete workflow you can reuse or automate.

Prompt — copy & paste

system / user prompt

We are defining a Claude agent workflow. For the goal I describe:
1. Name the agent and the trigger (e.g. "on PR open", "when I say /summary").
2. List 3–6 steps in order. Each step: input, action (read/call/write), output.
3. Specify the final output format (e.g. comment, file, Slack message).
4. Note any tools or MCP servers needed (e.g. GitHub, Notion).
If the workflow touches code, say which repo and branch. Prefer idempotent steps so the agent can retry safely.

▸ Expand — tips & usage

Use with Claude Code and MCP: once the workflow is defined, you can implement it as prompts + tools. Start with "Goal: [one sentence]".

↓ Related skills

MCP Integration Prompt Use when you want Claude Code to use an MCP server (e.g. GitHub, DB, Slack) in a safe and explicit way. Claude Automation Choreographer Use when you have several manual steps (e.g. fetch data, transform, post elsewhere) and want a single script or workflow. Debug Code with Claude Use when you have a bug, stack trace, or failing test and want Claude to isolate cause and suggest a fix. Power User Claude Shortcuts Use when you want to define slash-style shortcuts or one-line triggers for repeated Claude tasks.

← All skills