A lineup of AI skills
The Red line.
A set of AI skills for Claude Code — engineered to ship.
If you've only heard of Claude Code
An AI assistant that actually does the work.
Claude Code is a smart assistant that lives on your computer and doesn't just chat — it gets things done: makes images, voices texts, checks domains, tidies up your projects. “Skills” are its ready-made superpowers, one per task. No coding required — just ask in plain words.
Images & content
Generate visuals, icons, photos — content-gen.
Text to audio
Voice an article or make a podcast — audiocast.
Name a project
Invent and check free domains — domain-check.
Human-sounding text
Strip the “AI” tells from writing — humanizer.
Content
Images, audio, human-sounding text
Research & audit
Cited facts, SEO audit, domains
If you already get it
Orchestrators on top of a terminal agent.
Claude Code is a terminal agent. Skills are multi-agent pipelines on top of it: from idea to a site spec (redloft), panel-based plan review (plan-panel), stabilize + review over the diff (finalize), full task cycle to production (redwork), deep cited research (redresearch). Roles, a judge, self-improvement.
Idea to spec
brief→research→design→spec — redloft.
Plan review
architect+qa+judge panel — plan-panel.
Diff code review
stabilize + multi-role review — finalize.
Task to prod
implement→test→deploy autonomously — redwork.
A turnkey website
From an idea to a finished site — one orchestrator.
redloft is the flagship. It runs an idea through the full pipeline — brief → research → semantics → structure → content → design → spec — and hands you a ready build spec + a Claude Code prompt on a Next.js + Supabase turnkey base. It takes the pain of assembling a whole site from scratch: under the hood it calls the smaller skills itself, so you don't wire them up by hand.
Built from skills
Code without raw plans
Plan before code, review after — not by eye.
A multi-agent plan before the work is always raw, even on strong models. plan-panel (RedPlan) runs the plan through a panel of roles and a judge BEFORE code and catches gaps early. finalize stabilizes and does a multi-role diff review AFTER. redwork drives the whole cycle on autopilot. Three eternal coding pains — three fixes.
Built from skills
Websites
From an idea to a finished site
Design
Taste, animation, interface revision
design-reviseFor developersA revision loop for an existing block: taste → animation → audit.
Learn more
emil-design-engFor developersEmil Kowalski's philosophy of UI polish — the details that make software feel great.
Learn more
animateFor developersAnimation patterns and best practices for React/Next.js.
Learn more
design-motion-principlesFor developersMotion audit through three lenses: Kowalski, Krehel, Tompkins.
Learn more
page-design-pipelineFor developersCheckpoint-driven design of a page/component with approvals at each step.
Learn more
Code & quality
Planning, review, task autopilot
Research & audit
Cited facts, SEO audit, domains
Secrets under control
Stop scattering your keys everywhere.
API keys spread across .env files, notes and chats, leak into Claude Code history and commits, and can't be reused across projects. The secrets skill keeps everything in one place — 1Password — and injects the key straight into the process via op run: the value never lands in chat or in files. Useful in everyday work and in development alike.
See secretsOne place
Every key in 1Password, not ten scattered .env files.
Never leaks
The value is never printed to chat or written to files.
Reusable
One key across all projects, via an op:// reference.
Rotation
Change it in 1Password — it's picked up everywhere.
Technical standards
Local-first
Artifacts on your machine, nothing leaks.
1Password
Every key via op run, no plaintext.
Multi-agent
Roles + a judge with cross-examination.
Bilingual
RU and EN from one source.
Claude Code
Launched via slash commands.
Open source
The Red family — a public monorepo.