Prompts

Save prompts you use repeatedly, then compose them in an agent instead of rewriting them.

Two types

Type	Where it applies	Typical content
System	Role, boundaries, output format	“You are a code reviewer; report only reproducible issues”
User	User-side opening or constraint	“List your assumptions before answering”

An agent can select several of each; they are concatenated in list order.

Create a prompt

1. Open Prompts → New.
2. Title: by purpose, e.g. “strict JSON output”.
3. Type: system or user.
4. Content: supports template variables written as $variable or ${variable}.
5. Save.

Template variables

Write $name or ${name} and it is replaced at runtime. Useful for project names, output language and fixed field descriptions.

If a variable has no value, it is left as-is or handled per the UI hint; test in a session before relying on it.

Use in an agent

In the Agents editor, tick under System prompts or User prompts.

Order affects the result: put role definition first and output format last.

Writing advice

1. One prompt, one job, so it can be reused and replaced.
2. Make constraints checkable: format, length, required and forbidden content.
3. Never put keys or internal addresses in prompts; use Exchange Space instead.
4. Test in a small session before attaching it to a production agent.

FAQ

Edits do not take effect — confirm the agent references this prompt and that a new turn has started.

Variables are not replaced — check spelling and the $name / ${name} form.

Prompts conflict — split them and order them; keep only one version of conflicting constraints.