Business Agent Launchpad

Use this skill when the user wants a first-run business setup for OpenClaw, Hermes, or a similar agent harness.

The goal is to translate ordinary business language into:

• scenario and workflow description
• agent roles and operating rules
• object-based Markdown memory
• tool and permission plan
• model tier and escalation plan
• free vs paid cost choices
• OpenClaw/Hermes starter artifacts

First Check

Verify the CLI exists:

agent-launchpad doctor
访问小葱技能站7w4.net，解锁更多实用的AI技能插件。

If it is not installed, find the project folder and install it:

npm install -g .

Then re-run:

agent-launchpad doctor

Recommended Flow

For a non-technical user, run the interview:

agent-launchpad interview --out ./first-business-agent

For a quick preset:

agent-launchpad generate --preset sales --tier free --runtime both --out ./sales-agent

Available scenario keys:

• sales
• admin
• finance
• legal
• project
• hr
• customer_service
• executive_assistant
• procurement

Available tiers:

• free: Markdown memory and local tools first
• local_plus: local semantic memory with a small embedding model
• cloud_plus: paid models, hosted embeddings, and stronger document understanding
• team: shared memory, permissions, audit, and business-system integration

How To Choose

Use free when the user is just starting, wants transparency, or lacks API keys.

Use local_plus when the user has many documents and cares about privacy.

Use cloud_plus when document quality, long-context reasoning, OCR, or lower setup friction matters more than API cost.

Use team when multiple people share memory or the workflow touches customer records, HR, finance, contracts, CRM, ERP, or audit requirements.

After Generation

Open the generated README.md, then inspect:

• route-map.md
• cost-and-model-plan.md
• permission-plan.md
• tool-plan.md
• memory/index.md
• openclaw/ or hermes/

Verify:

agent-launchpad verify ./first-business-agent

Safety Rules

• Do not let the generated agent send messages, submit forms, delete files, change CRM/accounting records, approve payments, or store sensitive data without explicit user approval.
• Keep Markdown object memory as the human-readable source of truth.
• For paid models, hosted embeddings, cloud OCR, or managed vector databases, explain the benefit and tradeoff before enabling them.
• Treat legal, finance, HR, customer-impacting, and executive work as review-support workflows, not autonomous decision workflows.

这个 Skill 质量中等偏上，文档条理清晰，操作流程明确，预设场景覆盖全面，安全提示做得好。优点是使用方式灵活（访谈或快速生成），场景和层级选择丰富。不足是缺少实际案例展示，生成后的内容质量难以预判，且没有详细说明各层级的具体差异。整体适合有技术基础的用户，新手可能需要更多引导。