Agents

The inspector

A deeper view of one agent, in five tabs, Prompt, Skills, Workspace, Runs and Tasks, for when you want to check its work.

The inspector shows how an agent works: the instructions it works from, its files, each piece of work it did, and what it is doing now. You do not need it to use the app; it is there when you want to check an agent’s work.

Open it with Inspect in the agent’s Agent info. The back arrow returns to Agent info.

The inspector for Bill Payer on its Runs tab: each run with what started it, a summary of what it did, when, and how long it took.

It has five tabs:

Prompt: the agent’s instructions, the file AGENTS.md that it starts every session with, in full.
Skills: each of its skills, with its name and what it is for. A skill is a how-to the agent reads when a task needs it, kept as a file under skills/ in its workspace. A skill marked “Needs fixing” says what is wrong with it instead.
Workspace: every file the agent keeps, in folders, with each file’s size, when it changed and its first line. An agent files what it learns here and keeps refining it.
Runs: each piece of work the agent did, newest first. A run shows what started it (a message, a tag, or a task from the agent that created it), a summary of what it did, linked to the conversation it worked in, when, and the agent time it took.
Tasks: what the agent is doing now, live. Each task shows under the one that started it, such as a command under the reply it is part of, with its state (“running”, “waiting”). While the agent works, a line starting “Now:” says what it is doing. When it is idle, the tab says “Nothing running now.”

At the top of the Tasks tab, Computer shows the agent’s computer: whether it is running, paused or off and since when, how many commands are running on it, and, for a paused computer, how long its disk is kept. An agent starts a computer of its own when its work needs one, to run commands and tools. It pauses after a few idle minutes and is discarded when the agent’s session ends.