Anatomy of a great agent description
2 min
Four parts the wizard always wants: trigger, source, work, output. Name your providers explicitly.
The local SLM picks each step's behaviour from a closed catalogue (KIND_CATALOG). It's good at mapping verbs to kinds — but only when the description gives it concrete cues. A great description has four parts:
1. TRIGGER — when does the agent run? "Every weekday at 9am", "when a Slack message arrives", "on a webhook". 2. SOURCE — where does data come from? Name the provider: "my Outlook", "my Gmail", "the GitHub issues for org/repo". 3. WORK — what does the agent do with the data? "summarise key points", "extract amounts > $1000", "classify by priority". 4. OUTPUT — where does the result go? "email me", "post to #alerts in Slack", "create a Linear ticket".
Naming your providers (Outlook, Gmail, Slack) is the single highest-leverage thing you can do. The wizard's NER step picks them up + binds providers; without them the planner has to guess.
Live recipes need the desktop
This article is a static preview. The in-app Help sidecar inside Avery NXR can fire each step against your live project — install the desktop to use it interactively.