# Alt Text Sweep

> Walk every image in a documentation set and compose alt text that carries the same information the picture does, marking decorative art so it announces nothing at all. Use before an accessibility pass or a docs release. NOT for color contrast or focus order (use craft-review for those).

- Skill: `larrysequino/alt-text-sweep` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add larrysequino/alt-text-sweep`
- Raw SKILL.md: https://api.skillmd.com/api/skills/larrysequino/alt-text-sweep/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: LarrySequino (https://skillmd.com/u/larrysequino)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/larrysequino/alt-text-sweep

---


# Alt Text Sweep

Most alt text fails in one of two directions: it repeats the caption, or it
describes pixels. Neither helps. The test is whether someone who cannot see the
image ends up knowing what the sighted reader knows.

## Procedure

1. Inventory every image reference in the set, including ones inside partials and
   templates, which is where the untouched ones hide.
2. Sort each into one of three buckets: informative, decorative, or functional
   (an image inside a link or a button).
3. Informative gets the information, not the appearance. A bar chart's alt text
   is the finding the chart shows.
4. Decorative gets an empty attribute, deliberately, so a screen reader skips it
   instead of announcing a filename.
5. Functional gets the destination or the action, never the icon's name.
6. Re-read the page with the images removed. Anything that now reads as a gap is
   an image whose alt text is still wrong.

## Output

- The inventory, bucketed, with the current alt text beside the proposed one.
- The count of images that had no alt attribute at all.
- Any image whose meaning could not be determined from context, listed for the
  author rather than guessed at.