web-skill-atlas

Web Skill Atlas

Source-backed, reusable AI skills for designing, building, testing, debugging, and shipping modern web applications.

819 skills · 674 verified · 24/25 categories populated · 354 subcategories · 267 sources · updated 2026-08-27

Web Skill Atlas indexes reusable AI skills and agent workflows across public repositories, registries, tool ecosystems, and official documentation, then normalizes them around canonical sources.

This is not a generated prompt dump. Entries are sourced, classified, deduplicated, and linked to their public origin. The catalog is tool-independent and available as machine-readable data.

Browse · Example skills · Dataset · Verification · Contributing

If the atlas saves you time, consider starring the repository.

Browse

Browse the generated category pages or search the full dataset directly.

Example skills

A small cross-section of high-quality entries already in the dataset:

What counts as a skill?

A skill is a reusable instruction set, rule, workflow, or agent procedure for a concrete development task.

React is not a skill; a repeatable React performance procedure can be. Figma is not a skill; a reusable design-system audit can be. Collections are not split into invented variants just to increase the count.

Dataset

data/skills.json is the canonical dataset. data/index.min.json is the compact index. Generated indexes by category, tag, and framework are also available in data/.

curl -fsSL https://raw.githubusercontent.com/pneqnp-iswr/web-skill-atlas/main/data/skills.json \
  | jq '.[] | select(.tags | index("nextjs")) | {name, source_url}'

For coding agents, AGENTS.md and llms.txt point to the canonical files and repository conventions.

Verification

Canonical sources are preferred over registry mirrors, curated-list copies, and social posts. Forks and mirrors are not counted as separate skills by default. derived-workflow is a type, not a claim that an upstream project published that exact skill.

Quality scores measure practical usefulness, specificity, reproducibility, source quality, maintenance, uniqueness, and documentation. See docs/quality-scoring.md and docs/methodology.md.

Contributing

Found a missing skill or a bad entry? Use the issue forms or open a pull request.

For a new entry, provide at least:

• name and concise purpose;
• canonical source URL;
• suggested category;
• enough source detail to distinguish it from mirrors or nearby skills.

Before opening a PR:

python scripts/validate.py
python scripts/deduplicate.py
python scripts/generate.py
python scripts/validate_issue_forms.py
python scripts/scan_secrets.py
git diff --exit-code

The full contribution rules are in CONTRIBUTING.md.

Methodology

Discovery leads come from repositories, registries, official documentation, curated collections, and publicly indexed developer-community sources. Discovery is not verification: the atlas prefers the original source and records incomplete provenance explicitly.

Coverage is broad, not exhaustive. Search engines, GitHub indexing, social platforms, deleted repositories, and private or non-indexed sources all impose limits, and verification can lag behind upstream changes. See docs/research-log.md and docs/gaps.md.

Last dataset refresh: 2026-08-27.

Repository layout

.
├── data/          canonical dataset, source registry, generated indexes
├── categories/    generated category pages
├── docs/          methodology, taxonomy, scoring, research notes
├── scripts/       validation, generation, deduplication, link checks
├── assets/        repository presentation assets
└── .github/       CI, issue forms, pull-request template

See CHANGELOG.md for dataset and repository changes.

License

Web Skill Atlas code and original repository metadata are licensed under MIT where applicable. Third-party skills remain subject to their original licenses; indexing a source does not grant additional rights to its content. See NOTICE.