agentleFS
alphaXivalphaXiv/OpenResearch/.agents/skills/orx-accessibility/SKILL.md

orx-accessibility

Skill8k starsChanged 8 days agoMITView on GitHub

Make the OpenResearch app work by keyboard and assistive technology, and prove it without turning a screen reader on. Covers roles, names and states, ids, focus and Escape, keyboard shortcuts and the app menu, dialogs, menus and popovers through Base UI, live regions for status, long transcripts, reduced motion, contrast in both themes and on translucent surfaces, target sizes, zoom, and tests that assert on the accessibility tree. Use whenever you add or change a control, popup, dialog, menu, list, shortcut or focus behaviour in `ui/`, and whenever someone mentions screen readers, VoiceOver, NVDA, ARIA, keyboard navigation or WCAG. In this repo, use it in place of any general accessibility skill, including when reviewing someone else's change.

Adopted by

Steady: 0 new projects took it in the last 4 weeks, none the 4 weeks before.thin data

Week of Sep 13: 0 new projectsWeek of Sep 20: 0 new projectsWeek of Sep 27: 0 new projectsWeek of Oct 4: 0 new projectsWeek of Oct 11: 0 new projectsWeek of Oct 18: 0 new projectsWeek of Oct 25: 0 new projectsWeek of Nov 1: 0 new projects
Sep 2026Oct 2026Nov 2026
New projects taking a version of this file each week, over the 8 weeks our latest run read: only 0 in all, so this is a hint, not a trend.

Who's talking about alphaXiv/OpenResearch

The biggest 5 of 34 mentions linking to the repository that our crawl found. Each line is their own headline.

See all 34

Talked about on GitHub, Bluesky and Hacker News (34 times).

On GitHub · 22

17 more on GitHub

On Bluesky · 11

6 more on Bluesky

On Hacker News · 1

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    In e2e/,: use Playwright's getByRole, as the specs in e2e/tests/ do.

    line 13
  2. Very specific:

    Never hard-code an id: Every machine the user has opened stays mounted, so aria-labelledby, aria-describedby and htmlFor point at useId() …

    line 22

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Accessibility in OpenResearch

The tree is the test

Assert on roles, names, states, focus and live regions. Never turn VoiceOver, Narrator, NVDA or Orca on in the user's session or ask the user what they heard; a listening pass happens in a VM, when the user asks for one.

  • In unit tests, mount with render or renderApp from ui/tests/render.mjs and find things by what a screen reader exposes: find(`button[aria-label="${m.key()}"]`), [role=dialog], [aria-expanded=true]. find matches text content, so an icon button is found by its label. After press("Escape"), assert where document.activeElement is.
  • In e2e/, use Playwright's getByRole, as the specs in e2e/tests/ do.
  • Accessibility must not cost performance. The transcript mounts only its visible window, so it is an ARIA feed (role="feed", aria-setsize, aria-posinset, aria-busy; ui/src/components/ChatPanel.tsx) rather than every message in the DOM. A long list follows the same pattern.

Use the parts that already handle it

  • Dialogs come from useWorkspace().dialogs (orx-flows). Base UI moves focus in, traps it, and returns it on close; DialogTitle and DialogDescription name and describe the dialog. A dialog with no visible title takes aria-label.
  • Menus, popovers and selects use Base UI (@base-ui/react), which carries their roles, arrow keys and focus return. Never build one from divs and click handlers.
  • Buttons are Button and IconButton. An IconButton takes an aria-label from a Paraglide message naming the action, inside <Tooltip interactive> with the same message. A Tooltip without interactive is a focusable role="img" info icon: right for a lone hint glyph, wrong around a control.
  • Semantic elements first: button, a href, label, headings, lists. ARIA only where no element fits.
  • Never hard-code an id. Every machine the user has opened stays mounted, so aria-labelledby, aria-describedby and htmlFor point at useId() or useContext(WorkspaceDomContext).id(name).

Keyboard and focus

  • Every action has a keyboard path. Tab reaches every control in visual order, including chips and pills inside an editor; Enter and Space press; Escape closes the innermost popup and nothing else.
  • Escape is scoped. Wrap a surface that handles Escape in EscapeScope (ui/src/components/EscapeScope.tsx), which gives it the key exclusively while it is open.
  • App shortcuts live in appHotkeys (ui/src/hotkeys.ts), each named by a message. ui/src/appMenu.ts builds native menu items from those entries with command(appHotkeys[id]), added there by hand. Check a new chord against the existing ones and the ones each platform reserves, such as quit, close window and settings.
  • Focus is visible. Primitives draw the ring; a custom focusable copies theirs: focus-visible:outline-2 focus-visible:outline-solid focus-visible:outline-focus focus-visible:outline-offset-2. The ring wraps the control itself, never the row around it.
  • Focus goes where the next action is. A surface opened to be typed into focuses its input, as a new chat focuses the composer (useComposerFocus), and a closed popup or dialog returns focus to its trigger (Base UI). Follow the nearest existing mover; add a new one only with a reason.

Names, states and announcements

  • A name is what a sighted user reads, unique among its siblings, and it names the action, not the icon ("Close tab", never "X").
  • Expose every state the control has: aria-pressed for toggles, aria-expanded and aria-haspopup for anything that opens, aria-selected and aria-current for the current item, disabled with the reason visible nearby.
  • Status that changes without focus moving is announced through a live region (role="status", or role="alert" for errors): a turn finishing, a run failing, a copy confirming. Drawn but silent is a bug.
  • Don't announce noise. Streaming tokens and log lines don't go into a live region; their completion does.

Visual floor

  • Contrast: 4.5:1 for text, 3:1 for icons, control edges and focus rings, in light and dark, measured against what is actually behind the element: a translucent or blurred surface, a selected or hovered row, an image.
  • Targets are at least 24×24 px (IconButton small is 28).
  • Motion is optional: Tailwind motion goes behind motion-safe:, and a CSS animation in app.css gets a rule in its prefers-reduced-motion: reduce block, so it becomes an instant change.
  • Zoom works. The desktop app zooms with its zoom shortcuts in appHotkeys; check the change at 150% and 200% for clipping and overlap.
  • Colour is never the only signal. A failed run is red and says "Failed".

Checklist

  • role fits what it does, from a semantic element or a Base UI part
  • name is a Paraglide message naming the action, unique among siblings
  • every state the control has is exposed; ids come from useId() or WorkspaceDomContext
  • reachable by Tab in visual order; focus ring wraps the control; Escape backs out of the innermost popup
  • a popup takes focus and gives it back
  • status changes are announced; streaming text is not
  • contrast holds in both themes against what is behind it; targets are 24 px or more
  • motion is optional; zoom to 200% doesn't clip
  • a test asserts on the role and name of what changed

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.