FlaUI.Tool 0.1.6

flaui

Let your AI agent navigate your app, verify its own changes, and write the tests.

AI coding agents write your code. But when that code has a UI, the agent can't see what it built, can't check if the button works, and can't write a test for it. flaui closes that gap.

A stateless CLI built on FlaUI and Windows UI Automation (UIA3). Every command returns structured JSON. Every element selector is quality-rated. The agent drives it the same way it drives git or dotnet — from the terminal.

The workflow:

1. Agent writes code that changes a WPF form
2. Agent launches the app:          flaui session new --app "MyApp.exe"
3. Agent navigates to the form:     flaui elem find --name "Customer Details"
4. Agent verifies its changes work: flaui elem find --aid "EmailField"
                                    flaui elem type --id a1b2 --text "test@co.com"
                                    flaui elem find --name "Save"
                                    flaui elem click --id c3d4
5. Agent uses menus and keyboard:   flaui elem menu --path "File > Save As"
                                    flaui elem keys --keys "ctrl+s"
6. Agent reads the result:          flaui elem get-value --id a1b2
7. Agent captures visual proof:     flaui screenshot --output form-state.png
8. Agent exports the steps as test: flaui record export --out customer-form-test.json

---

Why flaui

• Agents can verify their own UI changes — the agent builds a feature, launches the app, and checks that buttons, fields, and workflows actually work
• Agents generate tests from real interactions — every click, type, and navigation is recorded and exported as structured test steps; the agent turns them into test cases
• Selector quality keeps tests reliable — every element is rated Stable, Acceptable, or Fragile; the agent knows which selectors will survive the next refactor
• Selector policies fail fast — set a quality floor per session; fragile selectors return exit code 2 instead of silently producing a flaky test
• Stateless and JSON-native — no persistent process, no shared memory; agents parse JSON natively, so there's zero glue code between the agent and the tool

---

Installation

Requires .NET 10 SDK or later. Windows only.

dotnet tool install --global FlaUI.Tool

---

Quick Start

# Agent launches the app it just modified
flaui session new --app "C:\Path\To\MyApp.exe" --selector-policy stable

# Agent explores the UI to understand the layout
flaui elem tree --depth 3

# Agent finds the field it added in the last commit
flaui elem find --aid "UsernameTextBox"

# Agent types into it to verify it works
flaui elem type --id a1b2c3d4 --text "testuser@example.com"

# Agent clicks submit to test the full flow
flaui elem find --name "Submit"
flaui elem click --id e5f6g7h8

# Agent reads back the result to confirm
flaui elem get-value --id a1b2c3d4

# Done — agent closes the app
flaui session end --close-app

---

Commands

All commands accept a global --session <path> option to specify the session file. If omitted, the most recent session file in the current directory is used.

session new

Launch an application and create a new session.

flaui session new --app <path> [--args <args>] [--selector-policy <policy>]

Option	Required	Description
--app	Yes	Path to the application executable
--args	No	Arguments to pass to the application
--selector-policy	No	Minimum selector quality: stable (default), acceptable, fragile

session attach

Attach to an already running application.

flaui session attach [--pid <pid>] [--name <name>] [--title <title>]

Option	Required	Description
--pid	One of three	Process ID to attach to
--name	One of three	Process name to attach to
--title	One of three	Window title to attach to

session status

Check if the session's process is still running and the window is valid.

flaui session status

Returns process alive state, window validity, element count, and recording status.

session end

End the session and optionally close the application.

flaui session end [--close-app] [--force]

Option	Required	Description
--close-app	No	Close the application when ending session
--force	No	Force-kill the application process (use with --close-app for unresponsive apps)

elem find

Find an element by one or more properties. Returns an element ID for use in subsequent commands.

flaui elem find [--aid <id>] [--name <name>] [--type <type>] [--class <class>] [--timeout <ms>] [--window <handle>] [--policy <policy>]

Option	Required	Description
--aid	No	AutomationId to find
--name	No	Element name to find
--type	No	ControlType to find
--class	No	ClassName to find
--timeout	No	Search timeout in milliseconds (default: 10000)
--window	No	Window handle (hex) to search in instead of the main window
--policy	No	Override session selector policy for this command (stable, acceptable, fragile)

Response includes selectorQuality and selectorStrategy so the agent knows how reliable the selector is.

elem tree

Dump the element tree as JSON.

flaui elem tree [--root <id>] [--depth <n>]

Option	Required	Description
--root	No	Element ID to use as tree root (default: main window)
--depth	No	Maximum tree depth (default: 3)

elem props

Get all properties of an element.

flaui elem props --id <id>

Returns AutomationId, Name, ControlType, ClassName, Bounds, IsEnabled, IsOffscreen, RuntimeId, HelpText, and AcceleratorKey.

elem click

Click an element.

flaui elem click --id <id> [--double] [--right]

Option	Required	Description
--id	Yes	Element ID
--double	No	Double click
--right	No	Right click

elem type

Type text into an element.

flaui elem type --id <id> --text <text>

Option	Required	Description
--id	Yes	Element ID
--text	Yes	Text to type

elem set-value

Set an element's value via the Value pattern.

flaui elem set-value --id <id> --value <value>

Option	Required	Description
--id	Yes	Element ID
--value	Yes	Value to set

elem select

Select an item in a combo box or list.

flaui elem select --id <id> --item <item>

Option	Required	Description
--id	Yes	Element ID
--item	Yes	Item to select

elem get-value

Get an element's current value.

flaui elem get-value --id <id> [--save <name>]

Option	Required	Description
--id	Yes	Element ID
--save	No	Save the value to a session variable for later use

elem get-state

Get an element's toggle/check state, enabled status, and visibility.

flaui elem get-state --id <id>

elem keys

Send keyboard input (key combinations or single keys).

flaui elem keys --keys <keys> [--id <id>] [--window <handle>]

Option	Required	Description
--keys	Yes	Key combination (e.g. ctrl+shift+s, tab, alt+f4)
--id	No	Element ID to focus before sending keys
--window	No	Window handle (hex) to target

Supported keys: ctrl, alt, shift, tab, escape/esc, enter/return, space, delete/del, backspace, up/down/left/right, home, end, pageup, pagedown, f1-f24, a-z, 0-9.

elem menu

Navigate and click a menu item by path.

flaui elem menu --path <path> [--window <handle>]

Option	Required	Description
--path	Yes	Menu path with > separators (e.g. "File > Save As")
--window	No	Window handle (hex) to target

elem scroll-into-view

Scroll an element into view using the UIA ScrollItem pattern.

flaui elem scroll-into-view --id <id> [--window <handle>]

Option	Required	Description
--id	Yes	Element ID
--window	No	Window handle (hex) to target

Returns scrolled: true if the element supported the ScrollItem pattern and was scrolled, or scrolled: false if the pattern is not supported. This is also called automatically by elem click, elem type, and other interaction commands.

window list

List all top-level windows for the attached application.

flaui window list

Returns an array of windows with handle (hex), title, isModal, className, and bounds.

window focus

Bring a window to the foreground.

flaui window focus --handle <handle>

Option	Required	Description
--handle	Yes	Window handle as hex string (from window list)

window close

Close a window by handle or title.

flaui window close [--handle <handle>] [--title <title>] [--force]

Option	Required	Description
--handle	One of two	Window handle as hex string
--title	One of two	Window title (partial, case-insensitive)
--force	No	Force-kill the window's process if graceful close fails

wait

Wait for an element condition to be met.

flaui wait --aid <id> --timeout <ms> [--value <value>] [--state <state>]

Option	Required	Description
--aid	Yes	AutomationId of the element to wait for
--timeout	Yes	Timeout in milliseconds
--value	No	Wait until element has this value
--state	No	Wait for state: hidden, visible, enabled

record start

Begin recording interactions. All subsequent elem commands are recorded as steps.

flaui record start

record stop

Stop recording.

flaui record stop

record drop

Drop the last recorded step.

flaui record drop

record keep

Mark the last recorded step as kept.

flaui record keep

record list

List all recorded steps with their sequence numbers, commands, and targets.

flaui record list

record export

Export recorded steps to a JSON file.

flaui record export --out <path>

Option	Required	Description
--out	Yes	Output file path

audit

Audit selector quality across all elements or recorded steps.

flaui audit [--window <id>] [--recording] [--out <path>]

Option	Required	Description
--window	No	Element ID to scope the audit to
--recording	No	Audit recorded steps instead of live elements
--out	No	Write audit report to file

Returns total elements, AutomationId coverage, selector quality distribution, and a list of issues (interactive controls with weak selectors).

screenshot

Capture a screenshot of a window or element and save it to a file.

flaui screenshot --output <path> [--id <id>] [--window <handle>]

Option	Required	Description
--output	Yes	Output file path. Format inferred from extension (.png, .bmp, .jpg)
--id	No	Element ID — captures that element's bounding rectangle
--window	No	Window handle (hex) — captures that window instead of the main window

If neither --id nor --window is given, captures the main window from the session.

Response:

{
  "success": true,
  "message": "Screenshot saved.",
  "outputPath": "C:/absolute/path/to/screenshot.png",
  "width": 1024,
  "height": 768
}

batch

Execute multiple commands in a single session (one process attach, shared element state).

flaui batch --file <path> [--continue-on-error]
flaui batch --steps '<json>' [--continue-on-error]

Option	Required	Description
--file	One of two	Path to a JSON file with batch steps
--steps	One of two	Inline JSON string with batch steps
--continue-on-error	No	Continue executing remaining steps after a failure

Input format:

{
  "steps": [
    {"cmd": "elem find", "args": {"aid": "SubmitButton"}},
    {"cmd": "elem click", "args": {"id": "$prev.elementId"}},
    {"cmd": "elem keys", "args": {"keys": "ctrl+s"}},
    {"cmd": "elem menu", "args": {"path": "File > Save As"}}
  ]
}

Use $prev.field to reference the previous step's result, or $steps[N].field to reference step N's result.

Supported commands: elem find, elem click, elem type, elem select, elem set-value, elem get-value, elem get-state, elem keys, elem menu, elem scroll-into-view, window list, window focus, window close, screenshot.

---

Selector Quality System

Every time flaui resolves an element, it evaluates the selector and assigns a quality rating:

Rating	What It Means	Agent Action
Stable	Unique AutomationId — the gold standard	Use with confidence
Acceptable	Resolved by Name + ControlType composite	Use, but note the risk
Fragile	Resolved by position, index, or volatile property	Flag for review or find a better selector
Unresolvable	Element not found	Retry, adjust approach, or report failure

Selector Policy

flaui session new --app "MyApp.exe" --selector-policy stable

With policy set to stable, any elem find that resolves below that threshold returns exit code 2 (policy violation). The agent knows immediately that the selector isn't reliable enough for production tests.

---

Test Generation from Real Interactions

The agent doesn't write tests from imagination. It writes code, runs the app, interacts with it, and turns those real interactions into tests.

# Agent finished implementing a login feature — now it verifies and records
flaui session new --app "MyApp.exe"
flaui record start

# Agent navigates the UI it just built
flaui elem find --name "Username"
flaui elem type --id <id> --text "admin"
flaui elem find --name "Password"
flaui elem type --id <id> --text "secret"
flaui elem find --name "Login"
flaui elem click --id <id>

# Agent checks the result
flaui elem find --aid "WelcomeLabel"
flaui elem get-value --id <id>

# Agent exports the verified steps as a test definition
flaui record stop
flaui record export --out login-test.json

# Agent audits selector quality to ensure test stability
flaui audit --recording

Every recorded step includes element selectors, quality ratings, timestamps, and parameters. The agent uses this structured JSON to generate test cases in whatever framework the project uses — NUnit, xUnit, MSTest, or a custom harness.

---

Exit Codes

Code	Meaning
0	Success
1	General error
2	Selector policy violation — element found but selector quality below policy threshold
3	Element unresolvable — element not found within timeout

Agents use exit codes for control flow. A non-zero exit tells the agent exactly what went wrong without parsing error messages.

---

Known Limitations & Workarounds

1. Elements inside collapsed WPF Expanders are not found

elem find returns "not found" for elements inside collapsed Expanders. This is correct UIA behavior — collapsed containers don't expose their children to the automation tree.

Workaround: First expand the Expander (find it, then elem click to expand), then search for child elements.

2. elem tree returns no children for WPF ComboBox

WPF ComboBoxes use virtualized item panels. Items only appear in the UIA tree when the dropdown is open and items are scrolled into view.

Workaround: Use elem select --id <id> --item "ItemName" to select items by name — this command expands the dropdown internally. To list items, open the dropdown first with elem click, then run elem tree.

---

Architecture

┌─────────────────────────────┐
│  AI Coding Agent            │
│  (Claude Code, Cursor, etc) │
└─────────┬───────────────────┘
          │ shell commands + JSON
          ▼
┌─────────────────────────────┐
│  flaui CLI                  │
│  stateless, per-command     │
└─────────┬───────────────────┘
          │ UI Automation (UIA3)
          ▼
┌─────────────────────────────┐
│  Windows Desktop App        │
│  (WPF, WinForms, Win32)    │
└─────────────────────────────┘

No SDK integration. No library imports. The agent calls flaui the same way it calls git or dotnet — as a CLI tool.

---

Building from Source

git clone https://github.com/kodroi/FlaUI.Cli.git
cd FlaUI.Cli
dotnet build src/FlaUI.Cli/FlaUI.Cli.csproj

---

License

MIT