igloude/ds-skills

README preview

# ds-skills

[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE.md)
[![Agent Skills](https://img.shields.io/badge/Agent%20Skills-compatible-8A2BE2)](https://code.claude.com/docs/en/skills)
[![Claude Code plugin](https://img.shields.io/badge/Claude%20Code-plugin-d97757)](#install)

Police large volumes of AI-generated frontend work for design-system adherence, plan features against the system before they're built, and audit the design system itself so it *can* be policed.

The idea: generation is cheap, the bottleneck is judgment. An expensive model spends its intelligence on judgment — is this a violation, how severe under this repo's policy, what exactly changes — and encodes that judgment into reviews, maps, and plans precise enough for the generating agent or a cheaper executor to apply. The skills are **report-only**: they never change a line of code, and every artifact they write survives a **cold read** — a reader with zero session context can act on the file alone. Readiness multiplies: every gap `ds-doctor` closes in the rulebook reduces violations at generation time *and* false positives at review time, across every agent and every branch.

All four skills are user-invoked (`/ds`, `/ds-doctor`, `/ds-plan`, `/ds-drift`) — they are deliberate, expensive operations, not something an agent should wander into, and their descriptions cost your sessions no permanent context. `/ds` is the router: it holds the diagram below and points you at the right sibling.

```
/ds-doctor    →  audits the DS itself, writes ds/MANIFEST.md
                  └→  everything below reads it

/ds-plan      →  analyzes feature plan, identifies gaps in DS
                  ├→ Covered / Composable  →  exact props, to the generators
                  └→ Extension / Net-new   →  DS work items inform the DS backlog

View full repository on GitHub →