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

# Test it before you trust it: checking work built on Athena

> Test any Skill, Agent, AOP, Automation, or app in Athena with saved examples, dry runs, and version previews before you publish changes to your team.

> Athena gives you draft testing, saved examples, dry runs, and version previews so you can check your work before people rely on it. You can test every part of Athena using the same method: write down what a good result looks like, try your work on real examples (including the hard ones), and replay those examples before you publish changes.

## What should I check?

Write down three to five checks that define a good result before you test anything. For a freight dispute review, your checks might be:

* The carrier and dispute ID come first.
* The billed amount is compared with the contract rate.
* Each piece of evidence is named by document.
* The review ends with one recommended action.

Next, collect five to ten real examples. Include hard cases that trip people up, such as a missing document, an unusual contract, or an amount right at a limit. Whenever something goes wrong later, add that case to your examples so you can test against it next time.

![A loop of four steps for every part: Skill, Agent, AOP, Automation and App. Write down what good looks like, as three to five checks. Try it on real examples, including the hard ones. Publish once the examples pass. Something went wrong? Add it as a new example. A bronze arrow back to trying it reads run the examples again before each publish. Footer: passing checks can still miss things.](https://mintcdn.com/athenaintelligence-e46bc9d3/U4e92C1IcbO5Ytjz/images/build-guides/test-before-you-trust-1.webp?fit=max\&auto=format\&n=U4e92C1IcbO5Ytjz\&q=85\&s=017ff3ce607376bc2697cd7f7352b81f)

## How do I test a Skill?

You can test a Skill right in chat by starting a new chat, mentioning the Skill, and running it on your examples. Check each answer against your list. When you fix and republish the Skill, start a fresh chat to test the update, since an open chat may keep using the earlier version.

To check a Skill against saved examples, put it inside an AOP (Agent Operating Procedure) and use the AOP testing tools below, since a Skill has no test button of its own.

## How do I test an Agent?

Athena lets you test Agent edits in a private draft before anyone else sees them. Open the Agent and choose **Test** to start a chat using the current draft, then try your examples. When the draft handles them well, choose **Publish** so your team's channels, Automations, and AOPs use the new version.

If your setup includes **Saved conversations**, Athena can replay past chats to check your updates. Store finished conversations along with their checks, and when you release the Agent in a Bundle with a proof step, Athena replays those conversations before the Bundle goes out and flags any replies that got worse. Use this replay to check the Agent's text reply, and test tool calls or actions separately.

## How do I test an AOP?

Athena gives AOPs built-in draft testing, saved examples, and automated grading against your checks:

1. Choose **Test run** and fill in the input form. Athena runs your draft and shows you any action that would change something, such as sending an email or updating a record, without carrying it out.
2. Under a finished run, choose **Looks right** or **Save as example** to keep it as an example.
3. Write your checks under **What good looks like**.
4. Under **Examples**, choose **Replay all**. By default, Athena replays up to eight of the newest examples on the draft and on the published version, if there is one. Athena grades each result against your checks and gives a reason for each pass or fail.
5. Choose **Publish**. Scheduled and triggered runs use only the published version, so your edits stay out of live work until you publish. If a new version causes trouble, open **History** to publish an earlier version again.

Select **Include test runs** whenever you want to see test runs in the run list, where Athena hides them by default. If a live run fails, open it under **Runs** and choose **Retry This Run**.

For a hands-on walkthrough, see [Build and trigger the exceptions AOP](https://resources.athenaintel.com/docs/athena-101/15-exceptions-aop.md) and [Approvals, failures and rollback](https://resources.athenaintel.com/docs/athena-101/16-approvals-failures.md).

## How do I test an Automation?

* Preview how your trigger behaves before running the Automation. **What would fire now?** takes a sample event and shows whether the trigger would start a run and with what inputs, while **Replay past events** checks real past events against your draft trigger.
* Choose **Dry run** to run the published version while Athena captures its changes instead of making them.
* Use **Dry run** when you want to avoid live actions, because both **Run now** and the **Test run** button on the Definition tab start a real run of the published version.
* Add an approval step so Athena pauses the run until someone approves it. See [Automation approvals](https://resources.athenaintel.com/docs/pillars/applications/automation-approvals.md).
* Set a spending budget per run and per day to cap costs. See [Automation budgets](https://resources.athenaintel.com/docs/pillars/applications/automation-budgets.md).
* Add a Judge step to have AI grade a result against your instructions during a run, so later steps can act on its verdict.
* Review each run's record to see the version, trigger, inputs, output, and cost, along with the inputs, outputs, and errors for every step. Use **Redrive** to restart a failed run from the first unfinished step or from a step you choose. See [Automation run history](https://resources.athenaintel.com/docs/pillars/applications/automation-run-history.md).

## How do I test an app?

* Ask Athena on the Computer to run the app's tests and checks. For a screen change, Athena can also test in a browser by planning up to four checks with pass rules, clicking through the app, and saving a recording with the results.
* Stage a new version to preview it without changing the live link, then open it with **Open version preview in new tab** or share it with **Copy preview link**. If the app is connected to GitHub, each pull request (a proposed code change) can get its own preview. Previews run without the app's real Secrets.
* Point the `health_path` setting at a page that reads from the database so Athena checks that this path responds before a version goes live.
* Test database changes on a safe copy by creating a [database branch](https://resources.athenaintel.com/docs/pillars/applications/databases.md), running the change there, and checking the app against it.
* Use **Version History → Roll back to this** to republish an earlier version. See [Manage application releases](https://resources.athenaintel.com/docs/pillars/applications/manage-application-releases.md).

## What to plan for

Athena caps AOP test runs at 40 AI calls or \$2 by default, and holds back browser clicks and typing even when a browser step opens real websites. Plan for test runs and dry runs to use your budget, because they read real data and make real AI calls.

Have someone who does the work today read the first live results. Passing checks give you a strong starting point, while human review catches cases your examples missed or places where AI grading was wrong.

Athena checks budgets against recorded spending, then fails the run and pauses the Automation if spending passes a budget cap. Keep some headroom in your budget caps since spending can pass a limit before Athena records it.

Put approval steps before any action you cannot undo, such as sending an email or changing a database. Rolling back a version republishes earlier code or instructions, while completed actions stay done.

## Questions teams ask

> [accordion] Is there one place for all my tests?
>
> Each part of Athena keeps its tests right where you build it: AOP examples and checks live in the AOP, saved conversations stay with the Agent, run history and Judge steps stay in the Automation, and tests and previews stay with the app.

> [accordion] Who should write the checks?
>
> The person who does the work today should write them, because they know what a good result looks like and which cases are hard.

> [accordion] How many examples do I need?
>
> Start with five to ten real examples, including hard cases, and add a new one every time something goes wrong.

> [accordion] Do test runs cost money?
>
> Yes. Athena holds back actions that would change something during a test run, while still reading real data and making real AI calls.

> [card] [See it work](https://athenaintel.com/demo/)
>
> Thirty minutes, your workflow, and you leave with the work product.


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