agentsclimarketplace

Accessibility pass

Skill protosphinx/sphinxstack/skills/accessibility-pass

A stack of skills for your AI agent, made for people starting from zero. 103 skills and 149 project briefs — load one into Codex, Claude Code, Copilot, Gemini, or Cursor and do the thing.

Install
npx -y skills add protosphinx/sphinxstack --skill accessibility-pass

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 16 days oldThe repository was created 16 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Audit a live site for accessibility and fix what turns up. Covers contrast, alt text, keyboard navigation, and form labels, tested with real tools. Use when they ask "is my site accessible", before they share a site widely, or after any skill that shipped pages.

SKILL.md

2.8 KB, as published. Nobody here has run it

accessibility-pass

Check someone's live site against how people actually use the web: on bright screens, without a mouse, with screen readers, with imperfect vision. Most fixes are small; the skill is finding them systematically instead of assuming the site is fine because it looks fine to its author.

Run the tools

  1. Run a Lighthouse accessibility audit (Chrome DevTools, same panel as make-it-fast) on every page. It catches the machine-checkable layer: contrast failures, missing alt, missing labels, missing lang. Save the score as the "before".
  2. axe DevTools, the free browser extension, on the worst page — it explains each finding better and catches things Lighthouse skips. One page is enough to learn the failure patterns.
  3. The tools stop at what machines can check, so add two human tests: put the phone at arm's length in daylight and try to read it, and do the keyboard test below.

Fix what turned up

Work through findings with them making the calls; each fix is a one-line explanation.

  • Contrast: every flagged text/background pair gets checked in the WebAIM contrast checker and nudged until it passes 4.5:1 (3:1 for large text). They pick the new shade — darkening the text usually keeps the design intact.
  • Alt text: every meaningful image gets a description of what it shows, written by them, since they know why the image is there. Purely decorative images get alt="" so screen readers skip them.
  • Keyboard: unplug the mouse. Tab through the whole page — every link and button reachable, in an order that makes sense, with a visible focus outline. If focus vanishes, someone styled outline: none; restore it or style a better one.
  • Labels and structure: every input has a <label> (forms-that-work does this from the start), the page has one h1 and ordered headings (shared work with seo-basics), links say where they go — "see the project" beats "click here".

Automated tools cannot prove that a site is accessible. Treat their findings as a repeatable first pass and record what was checked manually.

Done

  • Before/after reports saved, with every critical or serious automated finding resolved or documented with a reason
  • Full keyboard run-through works with visible focus throughout
  • Labels, heading order, image alternatives, and contrast checked manually
  • They can name the two or three things they will now do by default in everything they build next

That last point is the real product: the next site they build with build-website should not need this pass.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.