📖What it means

An outcome that is user facing: the agent actually carrying out a pre-defined activity for the user, such as triggering a task. An agent responding in a chat is NOT an action - the reply is conversation; the action is what gets done. The pre-defined set is the action catalogue, and anything outside it is answered with an offer to file an initiative. Typically an action is the response to a trigger a user makes through chat, internal or external: the person asks, the agent commits the action, deterministic code carries it out once and an action receipt records that it happened. Actions fold into tasks through agent chat: responding to a user is itself a task, its budget pays for the action, and so an action carries no task of its own. An action an agent can perform is one of its competencies, and how well it performs it is that competency's grade.

📖Easily confused with

[{"term": "Activity", "distinction": "An activity is a unit of work a droid runs; an action is an outcome a user sees. One action may be carried out by several activities."}, {"term": "Wake action", "distinction": "Something an agent may do on waking, triggered by its own wake and not by a user; not user facing."}, {"term": "Task", "distinction": "A task is the budgeted outcome an action folds into, not the action itself."}, {"term": "Chat reply", "distinction": "An agent answering in conversation. A reply may carry out an action, but replying is not itself one."}]

📖Nuance

Defined by Ninad, 2026-09-17, CLI-461. The estate also says 'wake action' for what an agent may do on waking (the conscious-action whitelist); that is not user facing and so is not an action in this sense. Budgeting through agent chat: Rulings Repo agent-resources #34 (supersedes #33).

📖Also called

- chat action
- user-facing action

📖Filed under

Category: governance.

Service: Governance.