Accessibility Auditor Pro

The free version audits one file in conversation. Pro scans your whole repo, computes real contrast ratios, blocks regressions in CI, and exports a prioritized fix backlog.

Free vs Pro

• Free audits one file per conversation; Pro scans entire directories with scripts/a11y_scan.py and JSON output
• Free quotes the 4.5:1 contrast rule; Pro computes the exact ratio and proposes a verified passing shade via scripts/contrast_check.py
• Free has no CI story; Pro ships scripts/a11y_gate.py with baseline diffing and exit-1 merge blocking
• Free leaves prioritization to chat; Pro exports a11y-backlog.csv with P0/P1/P2 priorities, duplicate counts, and effort points via scripts/backlog_export.py
• Free teaches the checklist; Pro adds references/wcag-decision-guide.md with the severity rubric and pipeline runbook

Upgrade Path

Upgrade when audits must be repeatable, automated, and enforced: multi-file repos, CI pipelines, sprint planning, conformance tracking. Nothing from the free setup is wasted: the WCAG knowledge, severity language, and report format carry over directly, and Pro findings reference the same criterion numbers your team already learned. One avoided post-release ADA complaint funds this skill for years.

Tags: accessibility, wcag, audit, ci-cd, frontend, compliance, testing

Version: 1.0.0

The Problem

accessibility-auditor has 9 installs — frontend teams clearly need WCAG help. But the free version audits one file at a time through chat. A real codebase has hundreds of components, shared form fields that propagate one missing label across forty pages, and status-badge grays that look fine but fail 4.5:1. Nobody will paste 200 files into a chat window, and nobody can eyeball whether #999 on white passes AA.

What You Get

• Repo-wide static scan via scripts/a11y_scan.py — walks src/, checks 12 violation patterns per line, tracks skipped heading levels, emits JSON with severity, WCAG criterion, exact fix, and effort points
• Real contrast math via scripts/contrast_check.py — computes the WCAG relative-luminance ratio for any fg/bg pair and suggests the nearest passing shade on failure
• CI regression gate via scripts/a11y_gate.py — diffs current findings against a checked-in baseline so only NEW Critical violations block the merge while legacy issues burn down
• Sprint-ready backlog via scripts/backlog_export.py — converts findings to a11y-backlog.csv with P0/P1/P2 priorities, site-wide duplicate counts, and summed effort points
• WCAG decision reference via references/wcag-decision-guide.md — severity rubric, contrast threshold table, and the 4-step scan-verify-gate-plan runbook
• Validated on a live fixture: 9 planted violations across lang, alt, buttons, links, labels, headings, iframes, live regions, and fonts all detected with correct severities

Why This Beats Prompting It Yourself

Asking a chatbot to check accessibility gives you the 4.5:1 rule quoted back and a guess about your grays. Pro runs the math: contrast_check.py computed #999 on white at 2.85 and proposed #747474 at 4.67, verified passing. Asking it to audit a repo gives you advice. Pro gives you a11y_scan.py finding every violation with file and line, a11y_gate.py blocking the next one in CI, and a CSV your sprint planner can import today.

Example Session

You: Audit our React dashboard in src/ for accessibility before Friday's release. It uses shared form components and custom status badges.

Agent: runs scripts/a11y_scan.py over src and reports 23 findings: 5 Critical, 11 Major, 7 Minor, 52 effort points total. The 5 P0 blockers name exact files and lines: 2 unlabeled form inputs (WCAG 3.3.2), 2 divs-as-buttons (WCAG 4.1.2), 1 page missing its lang attribute (WCAG 3.1.1).

Agent: the status badges flagged for contrast get verified with scripts/contrast_check.py instead of a guess. #999 on white computes to 2.85 against the required 4.5, a clear FAIL, and the tool proposes #747474, verified passing at 4.67.

Agent: runs scripts/backlog_export.py and hands you a11y-backlog.csv. 5 P0 items sit on top, 3 of them closed by a single shared FormField fix. P0 plus P1 totals 38 points, sized for one 2-week sprint. The free version would have quoted the checklist and asked you to paste files one by one. Pro scanned, measured, gated, and planned.

Use Cases

• Pre-release accessibility gate across a whole React, Vue, or plain-HTML repo without pasting files into chat
• Adding a two-line CI step that blocks PRs introducing new Critical violations while grandfathering legacy issues in a baseline
• Settling design debates with computed ratios instead of opinions, including batch palette checks via a pairs file
• Turning an audit into a sized sprint: duplicate counts reveal systemic single-component fixes, effort points size the work
• Tracking conformance over sprints with baseline diffs that show fixed-vs-new counts trending toward zero

Known Limitations

Static analysis only: the scanner catches structural patterns but cannot verify live keyboard tab order, screen-reader announcements, or API-rendered dynamic content. Requires Python 3.8 or newer, no dependencies. Output is an engineering audit input, not a legal ADA conformance attestation.