# Test before it goes live

> Runs the test parameters of an agent's script and functions, runs its behavior tests, and keeps a change staged until they pass.

Building an agent goes in order: build the agent, connect your systems, add functions, scripts and widgets, test it, take it live, then put it in your product. Highlighted: Test.

## Ask your coding agent

```
Run the test parameters for the invoice exceptions agent's script and its invoice comparison function, then write behavior tests for a matching invoice, a mismatched one and an unknown supplier. Run each test ten times and tell me which version they ran against.
```

Skill: [mindset-test-agent](https://docs.mindset.ai/skills/mindset-test-agent/SKILL.md)

## Test parameters

A script or a function keeps stored sets of inputs, its **test parameters**, and each set runs the whole path once. Give each path its own set, including one awkward case. A test run that writes needs an admin's go-ahead for its write list: your coding agent lists what the run will read and write, and waits.

Activating a version that puts new script text live is refused while the script calls an operation with no successful call since its last lapse, and publishing a function is gated the same way. Running the test parameters gives each operation that call. An admin can override per operation with **Go live untested**, recorded with their name.

## Behavior tests

A behavior test asks the agent a question many times and checks each answer against criteria, with a pass threshold. Every write is refused before it leaves, so a behavior test never changes another system. Behavior tests aren't a gate: activating a version doesn't run them, so run them before you decide.

Keep the tests in your repo with the agent's files, and accept the ones every run should check.

> **Info:** Behavior tests run with no person watching and without the agent's script, so the agent draws no charts or tables and gives its answer as text. Check a script's routing with its test parameters instead.

## What you decide

- The go-ahead for each test run that writes.
- Which behavior tests a new version must pass before someone activates it.

## Done when

- Every set of test parameters reads **Verified**.
- The accepted behavior tests pass on the staged version, and a person has activated it.

[Test behavior](https://docs.mindset.ai/docs/test-an-agents-behavior) covers comparing a change, such as a cheaper model, before you make it.