agent-builder
由 plgonzalezrx8 提供
Clawdbot Tools简介
Build high-performing OpenClaw agents end-to-end.
描述
该技能的功能
Guides you through designing an OpenClaw agent from scratch or refining an existing one. Runs a structured interview, then generates the full set of workspace files (SOUL.md, IDENTITY.md, AGENTS.md, USER.md, HEARTBEAT.md, and optionally MEMORY.md). Also covers guardrails, autonomy model, and acceptance tests.
Covers the full agent design lifecycle in one place, so you don't have to manually track which workspace files are required or remember what guardrails to include.
何时使用
- Setting up a new personal assistant agent with custom tone and memory
- Adding hard safety limits to an agent that keeps overreaching
- Defining group chat etiquette rules for a shared bot
- Iterating on an agent's autonomy level after observing failure modes
- Generating acceptance test prompts before deploying an agent
示例工作流
User asks: Build me a Telegram assistant agent that summarizes daily news and asks before sending anything
- 1
Ask clarifying questions: job statement, channels, autonomy level, hard prohibitions, memory needs, and tone
- 2
Generate IDENTITY.md, SOUL.md, AGENTS.md, USER.md, and HEARTBEAT.md from templates tailored to the answers
- 3
Apply guardrails checklist: ask-before-send rule, loop breaker, stop-on-error rule
- 4
Provide 5-10 acceptance test prompts to validate agent behavior
Complete OpenClaw workspace with all required files and a test suite ready for deployment
安装
clawhub install plgonzalezrx8/agent-builder运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。