cli-audit · git:20261112.07bb00b · 2026-11-12 · sha256 252ff866deccbd61

cli-audit git:20261112.07bb00bA

Immutable. This exact content never changes and is served at /api/v1/blob/252ff866deccbd61.

No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_ykfyxtb4z54c6oaa/252ff866deccbd611ea269ead3b9cfaaadb7d079bca000f760a5b0f53dfb67d3.json

Then /plugin install cli-audit@markdownregistry-ykfyxtb4z54c6oaa in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.

---
name: cli-audit
description: Audit live URLs using Playwright and Axe combined with WCAG skills
---

# CLI Audit Skill

Use this skill when asked to audit a live URL from Claude CLI.

Respect robots directives and site terms of service for every live-site audit request.

## Required Execution

When the user asks to audit a page URL, run:

```bash
npx --yes --package=github:mgifford/accessibility-skills cli-audit-inspect <URL>
```

This invocation is portable for Claude CLI users because it pulls the
inspection command from this repository package instead of requiring a local
`scripts/` file in the target project.

## Be Polite With Live Sites

When auditing live URLs, keep behavior respectful and low impact:

- Prefer single-page or small-batch checks first.
- Avoid rapid repeated scans against production pages.
- If Cloudflare/WAF blocks requests, do not brute-force retries.
- Recommend owner-approved alternatives: staging URLs, allowlisting, or manual checks.

## Required Parsing

Parse the emitted JSON and evaluate:

1. `axe.violations` for automated WCAG and best-practice failures.
2. `axe.incomplete` for potential violations that need manual confirmation.
3. `accessibilityTree` for behavioral issues that automated rules miss, including:
   - Missing or unclear names for interactive elements
   - Unexpected roles or role mismatches
   - Landmark and heading structure problems
   - State communication gaps (`expanded`, `checked`, `selected`, disabled state)
   - Dialog and menu semantics that can indicate focus-management risk

## Required Evaluation Method

Map findings to this repository's guidance before reporting:

- Start with `skills/ACCESSIBILITY-general/SKILL.md`.
- Then apply relevant topic skills based on observed problems (for example:
  `forms`, `keyboard`, `navigation`, `color-contrast`, `aria-live-regions`,
  `image-alt-text`, `tables`, `svg`, `touch-pointer`, `tooltips`).

## Reporting Requirements

For each issue reported:

1. Identify source: `axe` or `accessibilityTree`.
2. Include affected selector/path from axe nodes when available.
3. Provide impact summary and severity using the relevant skill scale.
4. Cite WCAG 2.2 success criteria when determinable, and cite the source of
   the mapping (the axe rule's documented mapping, or the specific skill
   rule) — do not invent a WCAG mapping this audit's evidence does not support.
5. Provide a concrete remediation recommendation tied to a skill rule.
6. Preserve the tool's raw outcome (`axe.violations` vs. `axe.incomplete`, or
   the accessibility-tree evidence) rather than collapsing it into a single
   pass/fail label. Playwright MCP refs and browser AX/DOM node IDs are
   session metadata, not tracker IDs or fingerprint inputs.
7. Report every `axe.violations` and `accessibilityTree` finding from this
   audit as `evidence_status: automated-indicator` with `handling: review`
   — this audit alone does not constitute the human confirmation needed for
   `confirmed-standards-failure`, `confirmed-user-facing-barrier`, or
   `handling: report`/`suppress`. See
   [Accessibility Finding Tracking: Policy Classification](https://mgifford.github.io/ACCESSIBILITY.md/examples/ACCESSIBILITY_FINDING_TRACKING.html#policy-classification).

If no issues are found, report residual risk clearly:

- Automated scans can miss interaction, timing, focus-flow, and assistive-tech
  behavior issues.
- Recommend manual keyboard and screen-reader validation using
  `skills/manual-testing/SKILL.md`.