If your tool does what a standard tool does, give it the standard name and fields. OrgRelay then routes it to the right department agent automatically, and the AI handles it the same way it handles every other connector.
Naming follows the rule in the connector spec: domain.action, lower case. The full input and output JSON Schemas are in spec/standard-tools.json in the developer kit, which you can download from the templates page.
Department agents in OrgRelay only see the tools whose domain they cover. The AI Chief, which answers general questions, sees every tool the person is allowed to use. With standard names the routing happens without configuration (OrgRelay 1.4.0 or later):
Agent
Domains it receives
Sales
orders, customers, products, reports
Support
tickets, orders, customers, products
Finance
finance, reports, orders
HR
hr, tasks
Procurement
purchase, inventory, products
Operations
inventory, orders, logistics, purchase, products, seo, content, ads, reports
The roles themselves, and what each agent is for, are described for business readers on orgrelay.com's agents page.
A second benefit: when a company connects two systems that both have orders.search, for example a WooCommerce store and an Amazon seller account, the AI can ask both the same way and compare the answers.
items[request_id, employee, type, from, to, status]
Salary, bank and identity data are deliberately absent. If a customer really needs them, put them in a separate tool the owner can restrict, as the security requirements describe.
Do I have to implement every standard tool? No. Start with the read tools your users ask about most. A connector with only read tools is valid.
My system has a concept that is not listed. Use your own domain with the same rule, for example quality.inspections. Only routing to department agents depends on the standard names.
Can I return extra fields? Yes. The listed fields are a minimum.