agent-builder

What This Skill Does

The Agent Builder is the cornerstone tool for designing and deploying high-performance OpenClaw agents. It facilitates an end-to-end workflow, from the initial conceptualization of an agent's persona and operating rules to the automated generation of the required OpenClaw workspace directory. By adhering to the standardized OpenClaw architecture—including essential files like SOUL.md, IDENTITY.md, AGENTS.md, USER.md, and HEARTBEAT.md—this skill ensures that your agent is built on a solid, maintainable foundation. It doesn't just create files; it functions as a design partner that guides you through an iterative, interview-based process to define autonomy levels, tool usage strategies, and critical safety guardrails. Whether you are building from scratch or refining an existing agent, the skill provides surgical assistance to enhance behavior, improve reliability, and optimize the agent's interaction model.

Installation

To integrate this skill into your environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/plgonzalezrx8/agent-builder

Use Cases

• Rapid Prototyping: Quickly spin up new agent instances with pre-configured, best-practice workspace files.
• Agent Optimization: Refine existing agents by analyzing past failures, adjusting their autonomy levels, and updating their core directives.
• Guardrail Enforcement: Programmatically inject safety protocols into your agents, such as 'ask-before-destructive' or 'stop-on-error' commands.
• Team-based Agent Design: Standardize your organization’s agent library by using consistent, professional templates for all internal bot personas.

Example Prompts

1. "I want to build a new agent to manage my personal calendar and email. Can you start the interview process so we can define its autonomy and tone?"
2. "My existing research agent is too chatty and keeps hallucinating tool results. Let's iterate on its SOUL.md and AGENTS.md to make it more concise and strict with its findings."
3. "Generate a new workspace for an 'Executive Assistant' agent. It should be an Operator-level agent with strict prohibitions against sending messages without my explicit approval."

Tips & Limitations

To maximize the utility of the Agent Builder, always ensure your workspace references are up to date. The skill relies on 'references/openclaw-workspace.md' and 'references/templates.md'; if these are modified, the generated output will adapt accordingly. Treat the interview phase as a critical design period; providing detailed answers during this phase drastically reduces the need for later manual edits to the SOUL.md or AGENTS.md files. Note that this skill is designed for advanced users who understand the implications of different autonomy levels; always test generated agents with the provided acceptance scenarios before deploying them to live production environments or external communication channels.