convez.ai
Skill
MattKay02

skills

MattKay02/skills

Ver no GitHub
InicialPluginScore · de 0 a 100

Claude Code skills I've built to augment how I develop — reusable agentic dev capabilities.

Instalação

$npx convez-ai MattKay02/skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 3 semanas
Versão
—
Issues abertas
0
Contribuidores
2

README

A tradução para o português deste README ainda está na fila.

skills

Claude Code skills I've built to augment how I develop — reusable, agentic capabilities I reach for across my own projects.

A skill here is a self-contained instruction set (the Claude Code SKILL.md convention) that gives an AI coding agent a specific, repeatable capability: a procedure to follow, with the judgement and guardrails baked in. These are tools I build around AI, not just prompts I type into it — each one encodes a workflow I'd otherwise repeat by hand, so the agent does it consistently and verifies its own work.

This repo is the canonical home for the generic, personal ones. Project- or client-specific skills live with their projects.

Skills

ui-walk · UI Evidence

Drives the running app with Playwright through every surface and state of a feature — empty, mid-flow, success, collapsed/expanded — saving an ordered, numbered screenshot folder that reads like a story. Idempotent: it cleans up everything it creates, so re-running after fixes regenerates the same sequence for comparison.

Why: I kept hand-writing the same walk script before every UI review — ordered captures of every state, then feed the folder to page-audit. Third time was the skill. · Stack: Claude Code, Playwright

page-audit · UI Review

Audits a UI surface across seven axes and returns a prioritized P1/P2/P3 punch list — verifying every finding against the actual code (file:line citations) before reporting it. A review, not a redesign.

Why: I kept doing the same manual UI pass by eye — this makes it systematic and evidence-backed. · Stack: Claude Code, Playwright

emulator-verify · Mobile QA

Lets an agent confirm a Flutter app's UI on the Android emulator by capturing screenshots over adb and reading them back with vision — so it checks its own work between turns instead of asking me to look.

Why: Closing the loop on mobile UI changes without a human in the middle every time. · Stack: Claude Code, adb, Flutter

lighthouse-audit · Web Perf

Runs Lighthouse against a production build, takes the median of several mobile runs (single runs are noisy), and turns the result into an actionable punch list — category scores, Core Web Vitals, the exact failing audits, and LCP phase breakdown — with before/after support.

Why: I kept hand-running the same spin-up-preview → run-Lighthouse → diff-before/after loop on every perf pass — this makes it one repeatable, median-stable step. · Stack: Claude Code, Lighthouse, Chrome

This list is generated from skills.json — the source of truth. Don't edit it by hand; add a skill there and the README syncs itself.

Using a skill

Each skill is a folder containing a SKILL.md. To use one with Claude Code, copy the folder into your skills directory:

cp -r page-audit ~/.claude/skills/

Claude Code discovers it automatically — invoke it by asking for the capability ("audit this page") or as a slash command (/page-audit).

License

MIT — see LICENSE. Use them, fork them, adapt them.


Built by Matthew Kay · mgkcodes.com · github.com/MattKay02

Ver o original no GitHub

Bom para

Relacionados