enterprise-agent-builder

name	enterprise-agent-builder
description	Build tenant-scoped business agents, knowledge, skills, data-fetch scripts, and invite flows for enterprise deployments.
version	1.0.0
metadata	{"hermes":{"tags":["enterprise","agent-builder","business-agent","skills"],"category":"enterprise"}}

Enterprise Agent Builder

Use this skill when an administrator wants to create, configure, or improve a business agent for any kind of business.

Operating Model

You are still a normal Hermes agent. Use native Hermes tools, native skills, and normal tool-calling behavior. The difference is scope: you are building tenant-scoped business agents and tenant/agent-scoped enterprise skill packages.

Use enterprise_builder for enterprise mutations. Do not edit enterprise.db directly.

Build Flow

1. Understand the business: industry, audience, products or services, common customer questions, escalation rules, and tone.
2. Identify the agent contract:
   • role prompt
   • primary tasks
   • tone and style
   • operating instructions
   • escalation boundaries
   • business knowledge
3. Identify capabilities:
   • built-in native skills that should be visible to this business agent
   • tenant/agent-specific enterprise skills that need to be created
   • data-fetch scripts needed for private business data
   • cron or follow-up workflows
4. For vague requests, produce a draft or ask focused questions. Do not create anything yet.
5. Apply the setup with enterprise_builder only after the admin explicitly confirms the specific draft.

Draft-First Rule

For an initial message like "create an agent for my bakery business", do not call mutating enterprise_builder actions. That request identifies an industry, but it is missing business-specific details.

First ask for the most important missing information, or provide a clearly labeled starter draft and ask whether to apply it. Useful questions include:

• business name and location, if relevant
• products/services and menu or catalog details
• order, pickup, delivery, cancellation, and refund policies
• allergy, safety, legal, or compliance boundaries
• whether the agent should answer only FAQs or also fetch live data
• which users should be invited
• preferred tone and languages

Only call mutating actions such as create_agent, update_agent, create_agent_skill, set_skill_catalog, or create_invite after the admin confirms with language such as "apply this", "yes create it", "save this agent", or equivalent.

When calling a mutating enterprise_builder action after confirmation, set confirmed_by_admin: true. Never set it to true for vague initial requirements.

Data-Fetch Skills

When an admin describes a business database, create a tenant/agent-scoped enterprise skill package instead of a global skill.

The package should include:

• SKILL.md instructions explaining when to fetch data and how to summarize it.
• scripts/*.py files when executable fetching is required.
• parameterized queries or clearly marked placeholders, never string-concatenated SQL.
• required environment variables or credential names, but never secrets in source.
• tenant, agent, and current-user scoping rules.
• a short test command or expected input/output format.

For example, if the admin says order history is in order_history and profile data is in user, create a skill that explains:

• fetch the current authenticated user's profile by user id.
• fetch only that user's orders.
• never query another user's records unless an admin explicitly authorizes an admin-only workflow.
• if credentials or column names are missing, ask before creating runnable scripts.

Safety Rules

• Do not claim an agent, invite, skill, script, or allowlist was created until enterprise_builder returns success.
• Do not call mutating enterprise_builder actions until the admin has approved a specific draft.
• Do not store database passwords, API keys, or customer secrets in prompts, SKILL.md, scripts, or knowledge fields.
• Scripts should read credentials from environment variables or a future secure connector layer.
• Use least privilege: generated scripts should query only the tables and rows needed for the current user workflow.
• Make business knowledge explicit. If a policy is unknown, instruct the business agent to say what information is missing instead of inventing.

Useful enterprise_builder Actions

• status: inspect tenant, users, and agents.
• list_builtin_skills: inspect native skills that can be allowlisted for business agents.
• create_agent: create a business agent with prompt fields and knowledge after explicit admin approval.
• update_agent: revise an existing business agent after explicit admin approval.
• set_skill_catalog: make a native built-in skill visible or hidden for a business agent after explicit admin approval.
• create_agent_skill: create a tenant/agent-scoped enterprise skill package, including optional scripts/*.py, after explicit admin approval.
• create_invite: generate an invite for a user and assign accessible agents after explicit admin approval.

Response Style

When applying changes, summarize what was actually created:

• business agent name and id
• enabled built-in skills
• created enterprise skills and package paths
• generated invite link or code if requested
• missing setup work, especially credentials or schema details

Keep the response concise and operational.