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:
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:
axe.violations for automated WCAG and best-practice failures.
axe.incomplete for potential violations that need manual confirmation.
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:
- Identify source:
axe or accessibilityTree.
- Include affected selector/path from axe nodes when available.
- Provide impact summary and severity using the relevant skill scale.
- 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.
- Provide a concrete remediation recommendation tied to a skill rule.
- 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.
- 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.
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.
1---2name: cli-audit3description: Audit live URLs using Playwright and Axe combined with WCAG skills4---56# CLI Audit Skill78Use this skill when asked to audit a live URL from Claude CLI.910Respect robots directives and site terms of service for every live-site audit request.1112## Required Execution1314When the user asks to audit a page URL, run:1516```bash17npx --yes --package=github:mgifford/accessibility-skills cli-audit-inspect <URL>18```1920This invocation is portable for Claude CLI users because it pulls the21inspection command from this repository package instead of requiring a local22`scripts/` file in the target project.2324## Be Polite With Live Sites2526When auditing live URLs, keep behavior respectful and low impact:2728- Prefer single-page or small-batch checks first.29- Avoid rapid repeated scans against production pages.30- If Cloudflare/WAF blocks requests, do not brute-force retries.31- Recommend owner-approved alternatives: staging URLs, allowlisting, or manual checks.3233## Required Parsing3435Parse the emitted JSON and evaluate:36371. `axe.violations` for automated WCAG and best-practice failures.382. `axe.incomplete` for potential violations that need manual confirmation.393. `accessibilityTree` for behavioral issues that automated rules miss, including:40 - Missing or unclear names for interactive elements41 - Unexpected roles or role mismatches42 - Landmark and heading structure problems43 - State communication gaps (`expanded`, `checked`, `selected`, disabled state)44 - Dialog and menu semantics that can indicate focus-management risk4546## Required Evaluation Method4748Map findings to this repository's guidance before reporting:4950- Start with `skills/ACCESSIBILITY-general/SKILL.md`.51- Then apply relevant topic skills based on observed problems (for example:52 `forms`, `keyboard`, `navigation`, `color-contrast`, `aria-live-regions`,53 `image-alt-text`, `tables`, `svg`, `touch-pointer`, `tooltips`).5455## Reporting Requirements5657For each issue reported:58591. Identify source: `axe` or `accessibilityTree`.602. Include affected selector/path from axe nodes when available.613. Provide impact summary and severity using the relevant skill scale.624. Cite WCAG 2.2 success criteria when determinable, and cite the source of63 the mapping (the axe rule's documented mapping, or the specific skill64 rule) — do not invent a WCAG mapping this audit's evidence does not support.655. Provide a concrete remediation recommendation tied to a skill rule.666. Preserve the tool's raw outcome (`axe.violations` vs. `axe.incomplete`, or67 the accessibility-tree evidence) rather than collapsing it into a single68 pass/fail label.697. Report every `axe.violations` and `accessibilityTree` finding from this70 audit as `evidence_status: automated-indicator` with `handling: review`71 — this audit alone does not constitute the human confirmation needed for72 `confirmed-standards-failure`, `confirmed-user-facing-barrier`, or73 `handling: report`/`suppress`. See74 [Accessibility Finding Tracking: Policy Classification](https://mgifford.github.io/ACCESSIBILITY.md/examples/ACCESSIBILITY_FINDING_TRACKING.html#policy-classification).7576If no issues are found, report residual risk clearly:7778- Automated scans can miss interaction, timing, focus-flow, and assistive-tech79 behavior issues.80- Recommend manual keyboard and screen-reader validation using81 `skills/manual-testing/SKILL.md`.