Agent authoring
The Agent Architect's toolset for drafting an agent definition in conversation. Drafts are never saved on their own — a person saves from the editor.
| Tool | What it does |
|---|---|
read_definition | Reads the conversation's current Agent.md draft. |
write_definition | Writes the Agent.md draft to the conversation's draft buffer. |
validate_definition | Runs an Agent.md through the registry's validation and returns field-by-field errors. |
search_catalog | Searches the agents and skills the person can see, so a new agent reuses what exists. |
read_definition
Reads the conversation's current Agent.md draft.
How an agent gets it. Declare it in the agent's tools: list. Only does anything when the run is an Architect conversation; elsewhere it is inert.
Arguments
No arguments.
write_definition
Writes the Agent.md draft to the conversation's draft buffer. It is shown to the person as a draft card; nothing reaches the registry.
How an agent gets it. Declare it in the agent's tools: list. Only does anything when the run is an Architect conversation; elsewhere it is inert.
Arguments
| Argument | Type | Required | Meaning |
|---|---|---|---|
definition_md | string | yes |
validate_definition
Runs an Agent.md through the registry's validation and returns field-by-field errors.
How an agent gets it. Declare it in the agent's tools: list. Only does anything when the run is an Architect conversation; elsewhere it is inert.
Arguments
| Argument | Type | Required | Meaning |
|---|---|---|---|
definition_md | string | yes |
search_catalog
Searches the agents and skills the person can see, so a new agent reuses what exists.
How an agent gets it. Declare it in the agent's tools: list. Only does anything when the run is an Architect conversation; elsewhere it is inert.
Arguments
| Argument | Type | Required | Meaning |
|---|---|---|---|
query | string | yes |