ai-skills

A collection of portable Claude skills. Each top-level directory is a self-contained skill with its own SKILL.md, supporting references/, scripts/, and a packaged .zip ready to install.

Skills

wcag-accessibility

Enforces WCAG 2.2 Level AA accessibility on any frontend code Claude writes or reviews — HTML, CSS, JSX, TSX, Vue, Svelte, web components, Tailwind. Has two modes:

• Authoring — bakes a11y in from the first keystroke: semantic elements, label/for associations, contrast, keyboard operability, visible focus, 24×24 touch targets, ARIA only where native semantics fall short.
• Audit — /wcag-check runs a structured pass over existing code against the W3C quick reference and produces a findings report with concrete fixes.

Triggers on frontend output and on phrases like "accessible", "a11y", "WCAG", "screen reader", "keyboard navigation", "ARIA", "alt text", "focus".

opentofu

Writing, reviewing, and migrating OpenTofu (1.6–1.12) infrastructure code, with explicit coverage of where OpenTofu diverges from Terraform:

• Terraform→OpenTofu migration with the backup-first / plan-parity-gate workflow and rollback path
• The compatibility cliff: state encryption, provider for_each, -exclude, early evaluation, .tofu files and destroy = false are OpenTofu-only; ephemeral resources and the HCP cloud {} block don't work
• State/plan encryption recipes (PBKDF2/KMS, unencrypted-fallback migration, enforced = true)
• Production state practice: native S3 locking, directory-per-environment, state splitting

Compiled from OmniMem experience memories and graveyard entries rather than hand-written — see the skill README for the recompile workflow. Triggers on "OpenTofu", "tofu", ".tofu", "state encryption", "migrate from Terraform".

working-backwards

Guides a user through Amazon's Working Backwards method to pressure-test a new idea and produce a complete proposal package:

• The Five Customer Questions
• A one-pager
• A full PR/FAQ (future-dated press release + external and internal FAQs)
• Tenets and a success-metrics sheet
• An optional six-pager narrative memo

Runs as a guided interview, includes a review pass that strips weasel words and marketing jargon, and is deliberately tool-agnostic so it works on any AI agent that can converse and write files. Triggers on "Working Backwards", "PRFAQ", "press release first", "one-pager", "six-pager", "narrative memo", or any request to develop, validate, pitch, or scope a new product idea.

Layout

Each skill follows the same shape:

<skill-name>/
├── SKILL.md          # the skill itself — frontmatter + instructions
├── references/       # deep-dive material the skill loads on demand
├── scripts/          # optional helpers (Python stdlib only where possible)
├── assets/           # templates, examples (some skills)
└── <skill-name>.zip  # packaged skill, ready to install

Installing

Drop the .zip (or the unpacked directory) into your agent's skills location. For Claude Code that is typically ~/.claude/skills/.