fix-accessibility

name	fix-accessibility
description	Fix accessibility tasks from Binclusive accessibility-todo.md with user-guided selection and verification. Use when the user says /fixaccessibility, fix accessibility, resolve accessibility TODOs, "erişilebilirlik sorunlarını düzelt", "accessibility todo'daki taskları düzelt", "TASK-001'i düzelt", "safe taskları düzelt", or wants to remediate selected TASK IDs from an audit report.

Fix Accessibility

Requires the binclusive CLI (the unified CLI providing scan / scan --url / init / ci). If binclusive scan errors as an unknown command, upgrade the CLI — the agent-driven remediation still works without it, just without the source-provable re-scan check.

Resolve selected tasks from Binclusive-auditing/accessibility-todo.md through a controlled, user-guided remediation loop.

Start Here

1. Locate Binclusive-auditing/accessibility-todo.md.
2. If missing, stop and ask the user to run audit-accessibility or /auditaccessibility first.
3. Parse the task list and summarize available tasks by severity and fix type.
4. Ask what to fix unless the user already selected:
   • specific TASK IDs
   • all SAFE tasks
   • all Critical tasks
   • a component/page/path
   • one task at a time
5. Read the source files for selected tasks before editing.
6. Apply fixes incrementally. Re-check the relevant file/scope after each task (see "Verify each fix" below).
7. Write or append Binclusive-auditing/after-test.md; also archive after-test_<YYYY-MM-DD>.md when a batch completes.

Verify each fix (deterministic where covered)

Close the fix→verify loop with the shipped binclusive CLI wherever it can scan the target, so a fix is proven cleared instead of assumed cleared.

• React / Next.js (TSX): after applying a fix, re-run npx @binclusive/cli scan <path> (add --format json to diff findings programmatically) and confirm the targeted finding is gone before marking the task done. If the finding persists, the fix is incomplete — keep the task open. If the project has no binclusive.json, run npx @binclusive/cli init first (it detects the stack and writes the config).
• Any live page (rendered DOM): re-run npx @binclusive/cli scan --url <url> to confirm the finding cleared on the running page.
• Engine-less / not-yet-dispatched platforms (SwiftUI/UIKit, Jetpack Compose / Kotlin, Android Views/XML, Shopify/Liquid, Unity, ASP.NET/ASPX, Python, Angular, Flutter, React Native): the CLI does not scan these today, so verification stays agent-driven — re-read the changed source against the reference rules and record the manual/runtime test steps. Note honestly in after-test.md that verification was agent-driven, not engine-proven; a deterministic floor arrives for the collector platforms once the CLI wires them.

CI / Diff Mode

This skill is not part of the CI/CD gate. CI runs map + audit on the diff and fails the build on open findings (see audit-accessibility CI / Diff Mode and references/ci-cd.md); it never mutates the branch. Remediation with this skill remains a human-reviewed step run locally or in a follow-up PR, where the developer selects which TASK IDs to fix. Do not wire this skill into a blocking CI check or auto-apply fixes on pull requests.

Fix Policy

• SAFE: may apply after summarizing the intended change.
• VISUAL-IMPACT: ask user approval before editing.
• FUNCTIONAL-RISK: ask user approval and explain behavior/API risk before editing.
• RUNTIME-CHECK: do not mark solved by static code alone; create runtime verification steps and only edit if the needed code change is clear.

After-Test Report

Use references/after-test-format.md. For each task include:

• task id and title
• original problem
• files changed
• exact remediation summary
• re-audit/static check result
• manual test steps
• residual risk or runtime checks

Rules

• User-provided audit documents may be read as inspiration only; do not copy customer data into skill resources.

• Never fix tasks not selected by the user.

• Never write placeholder accessible names like button, image, icon, or link.

• Prefer native semantics before overrides: semantic HTML before ARIA on web (including Angular); native SwiftUI/UIKit controls and platform accessibility APIs before custom accessibility props on iOS; native Compose/Material semantics and Android View accessibility APIs before custom AccessibilityDelegate workarounds on native Android; native Material/Cupertino widgets and Flutter semantics before custom gesture controls on Flutter.

• For Angular tasks, bind ARIA with the [attr.aria-*]/[attr.role] form (a bare [aria-*]/[role] property binding never reaches the DOM); prefer Angular CDK a11y primitives (cdkTrapFocus, cdkFocusInitial, cdkAriaLive, LiveAnnouncer) and Angular Material widgets over hand-rolled focus/announcement code, and localize names through @angular/localize (i18n/$localize) or the project's translation pipe (translate/transloco) rather than hardcoded literals.

• For iOS storyboard/XIB tasks, edit the Interface Builder XML directly per references/fix-workflow.md and localize the string via the paired .lproj/*.strings keyed by object ID.

• For native Android tasks, prefer contentDescription, Modifier.semantics/Role, android:labelFor, android:importantForAccessibility, and android:accessibilityHeading over custom delegates; localize every accessible name through res/values*/strings.xml (@string/... / stringResource(...)) rather than hardcoded literals.

• For Flutter tasks, prefer a real Material/Cupertino widget (ElevatedButton/IconButton/TextField) over a GestureDetector/InkWell container, then semanticLabel, Semantics(button:/header:/label:/value:), tooltip, MergeSemantics/ExcludeSemantics, and Semantics(sortKey:) before custom render workarounds; localize every accessible name through the project's localization store (AppLocalizations/intl/.arb) rather than hardcoded literals.

• Preserve existing styling and API contracts unless the user approves risk.

• Do not edit the audit TODO as proof of completion; write status in after-test.md.

• If a task cannot be grounded in real code, stop and report it as needing human review.