> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tyba.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent stats

> How many approvals the agents asked for, how many you had to decide, and how long you took to decide.

TYBA was already recording two sets of data nobody looked at: every approval request — with the command, its risk, the decision, and the times of the request and its resolution — and every command run, with exit code and duration.

The question they answer is not "which command do I use most". It is **how much of your attention the agents cost**.

Open it from the command palette, under *Agent stats*.

<Note>
  **Read-only.** Nothing on this screen runs, re-runs or deletes. It is a picture of what already happened.
</Note>

## The cards

|                           |                                                                  |
| ------------------------- | ---------------------------------------------------------------- |
| **Approvals requested**   | Everything that went through the gate in the period              |
| **Auto-approved**         | Green risk, or something already cleared earlier in that session |
| **Decided by a human**    | Someone had to stop and look                                     |
| **Denied**                | You said no                                                      |
| **Median time to decide** | How long the agent sat waiting on you                            |

The median time is the number that changes decisions: if it grows, the agent is blocked longer than you think, and it is worth reviewing what keeps landing in the gate.

## The tables

**Commands the agent asks for most** — with each one's risk and how often it gets approved. A command asked for constantly and always approved is a candidate for a rule; one asked for constantly and sometimes denied is the one that deserves attention.

**Agent sessions** — how many commands ran, how many approvals passed through, and how long the session lasted.

## Filters

By period — **7 days**, **30 days** or **everything** — and by **repository**, because an agent working in a write-heavy repo behaves differently from one that only reads.

## See also

<CardGroup cols={2}>
  <Card title="Agents board" icon="squares-four" href="/en/agent/agents-board">
    Who needs you now, rather than what already happened.
  </Card>

  <Card title="Claude Code and Codex" icon="robot" href="/en/agent/claude-code-and-codex">
    Where the approval requests come from.
  </Card>
</CardGroup>