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

# Runs, findings and changes

> What an agent did, what it found, and how to undo a change it made in Shopify

## Runs

A run is one time an agent woke up and worked. An agent's runs are on its
**Runs** tab, and a deployment's on its own Runs tab.

Filter by **Store**, **Run status**, **Trigger** (Schedule, Event, Run now or
Chat) and what the run **Produced** (findings, changes or reports).

| Status | What it means |
| - | - |
| **Running** | Still working |
| **Ok** | Finished |
| **Nothing to do** | Woke, found nothing to work on, and stopped |
| **Waiting on you** | Paused for an answer from you |
| **Halted** | Stopped before finishing, for example when cancelled |
| **Error** | Failed. The run says why. *"turn budget reached"* means it used its whole budget; see [Budget per run](https://docs.glood.ai/platform/agent-tools.md#budget-per-run) |

Select a run to open it:

* **Why it ran**: the schedule, the event (a new order, a return request, a
  refund) or the person who started it.
* **What came out**: its findings, changes, reports, memories, assessments and
  alerts, or *nothing to report*.
* **Steps**: everything it did, in order: its thinking, each tool it called with
  **Show input and output**, and each finding it raised or updated.

The agent's thinking is kept for 90 days. After that the tool steps and their
results are still shown.

## Findings

A finding is something an agent recorded as true, with its evidence. Go to
**Agents → Findings** for the findings of every agent in the organization,
newest first, or pick one **Agent**. Filter by **Severity** (Critical, Warning,
Info) and **State**, or search.

| State | What it means |
| - | - |
| **Open** | Still true, as far as the agent knows |
| **Applied** | Someone recorded that they acted on it |
| **Dismissed** | Someone recorded that it does not need acting on, and why |
| **Resolved** | The condition went away |

Select a finding to read it:

| Part | What it holds |
| - | - |
| **Because** | Why the agent believes it |
| **What moved** and **Evidence** | The figures and facts behind it, and any insights it explains |
| **Ruled out** | Causes the agent checked and rejected, and why |
| **Recommend** | What the agent suggests you do |
| **History** | Who opened it, when it was first and last seen, how many times, and the episode before it if it came back |
| **Alerts sent** | Where an alert about it went |

An agent never raises the same finding twice while it is open. When it sees the
condition again it adds to the same finding, and the count of times grows.

### Applying or dismissing a finding

The dashboard shows findings but does not record a verdict on them. To mark one
**Applied** or **Dismissed**, ask your [AI assistant](https://docs.glood.ai/platform/using.md), for
example *"dismiss that finding, it was a planned price test"*. The reason you
give is kept on the finding.

## Changes

When an agent in **Act** mode changes something in Shopify, it is listed with a
way to undo it. Today that is the Fraud agent's tags and risk assessments on
orders.

Go to **Agents → Changes** for every store, or a store's **Changes** tab for one.
Each row shows the order, the action (**Added tags**, **Removed tags** or **Risk
assessment**), what changed, the reasons, the status, and the run that made it.

| Status | What it means |
| - | - |
| **Pending** | Being written to Shopify |
| **Applied** | Done |
| **Failed** | Shopify refused it. The row says why |
| **Reverted** | Undone |
| **Revert skipped** | Nothing was left to undo: the order no longer carries the change, or no longer exists |

### Revert a change

* **Revert** on a row undoes that one change.
* **Revert all changes in run** undoes everything one run changed on the store,
  newest first. Filter the list by run, or open the run, to find it.

Tags the agent added are removed; your own tags never are. Shopify cannot delete
a risk assessment, so reverting one adds a new assessment at no risk.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.