# Semantics Lens

> An accessibility audit for one Flutter screen. Paste the Dart (or drop a .dart file); the browser masks comments and
> strings, matches every widget call to its closing bracket and reads its named arguments, then finds icon buttons with
> no tooltip, custom tap targets with no label or button role, images neither labelled nor excluded, tap targets under
> 48 x 48 or shrink-wrapped, text scaling switched off or capped, fixed heights that clip enlarged text, literal colours
> under 4.5:1 (3:1 for large text and icons), tappable rows read in pieces, and Flutter web with no semantics tree.
> Metered runs write the audit and the Semantics-patched code. Built on the SkillSafe skills
> @flutter/flutter-improving-accessibility and @flutter/flutter-accessibility-audit, from flutter/skills
> (BSD-3-Clause), by the Flutter team. An independent tool, not affiliated with or endorsed by
> Google or the Flutter team.

https://semantics-lens.skillsafe.ai/

## What is free
- The scan: every finding with an id (S1, S2, ...), a severity (blocker, major, minor), the line, the widget, what the page saw and the fix.
- Contrast measured for every literal text or icon colour against the literal background that encloses it (Color(0x...), Color.fromARGB, Colors.* and their shades and alpha variants).
- The skill's checklist marked from the scan (pass, fix, check by hand), test/a11y_test.dart asserting androidTapTargetGuideline, iOSTapTargetGuideline, labeledTapTargetGuideline and textContrastGuideline for the screen's widget class, a findings CSV and everything as JSON.

## The metered lanes
Two tasks on gpt-terra, metered per run, routed by `task`. Input string fields: `task` (audit or patch), `screen`, `code`, `carry` (patch only: the audit's decisions) and `facts` (the page's scan).
- `audit` replies with `summary`, `findings` (confirmed ids with the barrier, the fix and a short code fix), `cleared` (ids that are not barriers, with why), `extra` (up to five barriers a scan cannot see, on real lines), `reading_order` and `caveats`.
- `patch` replies with `summary`, `code` (the whole patched file), `changes`, `kept`, `test` and `caveats`.
The page holds every reply to its scan: each finding id accounted for exactly once; for the patch, every widget class and the whole file kept, the test asserting all four guidelines, and the patched code re-scanned - a claimed fix the re-scan still finds, or a new finding the patch introduces, is flagged.

## What it cannot see
Colours from the theme, sizes set by layout, focus order and live announcements need the widget test, TalkBack or VoiceOver.

## Links
- API tutorial: https://semantics-lens.skillsafe.ai/api.html
- Source skills: https://skillsafe.ai/skill/@flutter/flutter-improving-accessibility and https://skillsafe.ai/skill/@flutter/flutter-accessibility-audit