# Setting up accessibility

**At the end:** Loadout, the agent it starts and the dashboard all following
the way you've said you read, and you knowing which parts of that have been
checked.

## Before you start

- Loadout installed. See [installing](installing.md).
- Nothing else. Loadout detects nothing about you or your machine: you turn this
  on yourself, and nothing is guessed.

## Steps

### 1. Choose a preset

A preset is a starting point. It sets a bundle of settings, and anything you set
yourself wins over it. Presets are named for what they reduce, not for a
condition, because you shouldn't have to declare a diagnosis to get shorter
sentences.

| Preset | What it sets |
|---|---|
| `screen-reader` | No redraws, ASCII glyphs, lists instead of tables, numbered menus, the text launcher, no motion, the terminal bell, one question at a time |
| `low-vision` | Sixteen colours, reduced motion, bold-only emphasis, a summary first |
| `colour-blind` | Sixteen colours, and no red-and-green pairs |
| `dyslexia` | Sentences under twenty words, paragraphs of four, bold-only emphasis, one word per thing, numbered steps, plain language, a summary first |
| `adhd` | One question at a time with "question 2 of 4", a summary first, concise answers, confirmation before anything irreversible, reduced motion, no redraws |
| `plain-language` | Plain words, every question explained, one word per thing, and why it is being asked |

### 2. Try it on one command first

```sh
loadout team list --accessible
```

The same list under the screen-reader profile, one labelled line per value.

```text
$ loadout team list --accessible
Accessible output is on (screen-reader). Change it with: loadout config set accessibility-preset <name>
bug-hunt  5 node(s), supervised
  Reproduces a bug without a person watching, proves its cause, fixes the proven
cause, and proves the fix holds.
dependency-sweep  4 node(s), supervised
  Updates dependencies one branch per bump, in parallel, and verifies each with 
the suite green before it is offered for merge.
docs-crew  4 node(s), supervised
  Finds where the documentation and the code disagree, fixes it in the voice the
docs already have, and follows the changed pages as a new reader would.
iterating-project  5 node(s), supervised
  Plans, implements, reviews and verifies in rounds until the goal is met, the 
budget is spent, or two rounds make no progress.
marketing-studio  5 node(s), supervised
  Turns a goal into a strategy, writes the pieces, edits every claim against the
facts, and stages each send for you. Nothing is sent in any mode; your own copy 
may allow a named channel.
product-company  template  17 node(s), supervised
  A shape to copy, not a team to run. An executive lead splits a goal across 
department leads, each with its own workers. Deliberately shallow, because every
level between you and the work costs a session and loses part of the brief.
release-crew  4 node(s), supervised
  Checks the tree is fit to release, writes the notes for somebody deciding 
whether to update, and tags. The push is a gate in manual and supervised runs; 
an autonomous run may push the tag.
system-watch  6 node(s), supervised
  Investigates a system problem, proves the cause, fixes it, and turns the fix 
into something the team keeps and reuses.

Run one with: loadout team run <team> "<goal>"
```

To try a particular preset, name it: `--accessible=dyslexia`.

You should see a first line naming the profile that's active. That line is how
you know it took. Under `screen-reader`, tables become one labelled line per
value and the output is ASCII only.

### 3. Or turn it on for this terminal

```sh
export LOADOUT_ACCESSIBLE=screen-reader
```

On Windows PowerShell:

```powershell
$env:LOADOUT_ACCESSIBLE = "screen-reader"
```

`1`, `true` and `yes` mean the screen-reader preset; `0`, `false` and `none`
mean off.

### 4. Turn it on for good

```sh
loadout config set accessibility-preset screen-reader
```

These three places are checked nearest first: the flag, then the environment
variable, then the setting. So a flag on one command always wins.

`NO_COLOR` is obeyed whatever the preset says, and switches off every escape
sequence, not only colour, because bold is an escape too.

### 5. Open the launcher

```sh
loadout
```

Under `screen-reader`, you should see a numbered text menu instead of the
full-screen launcher, answered by typing a number.


The text launcher is the tested path. Under the other presets you get the
full-screen launcher, drawn in the sixteen colours your terminal theme defines,
in ASCII where you asked for it, and without its opening animation where you
asked for less movement.

### 6. Check what the agent will be told

```sh
loadout loadout-cli --dry-run
```

Replace `loadout-cli` with your project's slug. The dry run says where the
compiled context went, and that context ends with a section saying how you
asked to be written to. It changes how things are said,
never what is done, and never what's quoted: a log line or an error is always
reproduced exactly.

When the session starts, Claude Code is started in its own screen-reader mode
where you asked for no redraws. Codex has its animations switched off; it has
no screen-reader mode, and Loadout says so at launch.

### 7. Open the dashboard

```sh
loadout dashboard
```

Under `screen-reader` or `low-vision`, or with colour set to none, you get the
plain page.

![The plain dashboard: the same four runs as a single list, each row giving the run's name, its state in words, and what it has cost.](../images/dashboard-list-plain.png)

A **Plain view** button is early in the tab order on every page, and the change
is announced in the page's live region. [Using the dashboard](dashboard.md)
covers the rest.

### 8. Turn on speech, only if you want it

Speech is off by default, and **stays off even under the `screen-reader`
preset**. To turn it on:

```sh
loadout config set show-speech screen-reader
```

With it on, the full-screen launcher says each row as the cursor lands on it,
the project's name first, so you can interrupt as soon as it's the wrong row.
The dashboard speaks its announcements too, but only when all four of these
hold: a screen reader answered rather than a system voice, `show-speech` is
`screen-reader`, the browser is on the same machine, and the request carries
the dashboard's token. The settings page says which one is missing. Nothing is
said that isn't also written on the page, and a line over 400 characters is
cut.

**Speech is built, and has been heard only through the Windows system voice,
never through a screen reader.** That's why it's off by default: offering it
unasked would be offering an unverified experience to exactly the people who
can't check it.

## What has been verified, and what hasn't

| | Status |
|---|---|
| Each preset produces the guidance, flags and settings it promises | Tested |
| Accessible output carries no escape sequences | Tested |
| Menus are numbered, and refuse a number outside the list | Tested |
| Dashboard: axe-core 4.10.2, nine states on both pages in all five themes | No violations |
| Dashboard text contrast, every text colour on every surface | 4.7:1 at worst in Daylight, 5.6:1 in the dark themes, 7.3:1 in High contrast, against 4.5:1 required |
| Dashboard accents, each on every surface of every theme | Pass, worst 4.8:1 |
| Dashboard at 320 pixels wide and at 200% zoom, every screen, both pages | No sideways scrolling |
| Dashboard speaking: the order in which it refuses | Tested against a stub, four mutations each failing its own test |
| Speech through the Windows system voice | The route answers |
| Speech through NVDA, or through `say` on macOS or speech-dispatcher on Linux | **Never heard** |
| Any screen reader, used with any of this | **Not tested** |
| A keyboard-only pass by a person | **Not recorded** |

If you use a screen reader, an issue saying what you heard is the most useful
thing you could send.

## If it went wrong

- **The first line doesn't name a profile.** Check for a flag or
  `LOADOUT_ACCESSIBLE` overriding the setting: the nearest one wins.
- **You hear nothing.** Speech is off unless you set `show-speech`. The
  dashboard's settings page names the condition that isn't met.

## What this doesn't do

- It doesn't detect anything. Nothing is turned on by guessing at your needs.
- It doesn't change either agent's colour theme, because which one you want
  can't be known from a profile.
- No dyslexia font and no readability score. Studies of dyslexia fonts find no
  gain, and a score measures sentence length, not understanding.

## Next

[Accessibility reference](../accessibility.md), for every setting on its own.