The kit, cell by cell.
13 skills, fully explained
Every public skill in the kit — its command, what it does, and how it
works under the hood. No black boxes: git clone the
source, run make, and read along.
workflow
From rough idea to verified merge — planning, publishing, and running entire features.
story-workflow
Runs large features as stories — an epic issue with one sub-feature workflow per slice, structured as a vertical tier bracket.
Analyzes once at epic level with maximum strictness, writes a bracket file of sub-features with dependencies, then launches each sub-feature as an isolated workflow that inherits the analysis. A watchdog resumes stalled slices until the story verifies complete.
brainstorm
Turns a rough feature idea into a validated spec and implementation plan before any code exists.
Explores intent and requirements through clarifying questions, challenges assumptions, then produces a spec plus a step-by-step plan. Optionally publishes the plan to Linear and sets up an isolated git worktree for the work.
writing-plans-linear
Writes implementation plans detailed enough for an engineer with zero context, and publishes them to Linear for team visibility.
Decomposes the spec into bite-sized tasks — files to touch, code to write, how to test each step — assuming the reader knows nothing. In publish mode it creates one Linear subtask per milestone and syncs progress as work lands.
orca-superpowers-workflow
The end-to-end feature pipeline: impact analysis, Linear issue, spec, plan, task DAG, execution with gates, and final verification in one flow.
Wraps the full superpowers workflow and automatically activates the Orca bridges at transition points — Linear sync, worktree creation, task DAG as external memory, and decision gates between phases. Replaces juggling five separate skills with one command.
design
Interfaces with intent — direction, motion, review, and translation from image to code.
frontend-design
Design guidance for UI that reads as intentional and distinctive — never templated defaults.
Approaches each brief like a design lead: ground the design in the subject's own world, make deliberate typography and palette choices, and take one justified aesthetic risk. Loaded whenever new UI is being shaped.
gpt-taste
Elite UX/UI and GSAP motion engineering rules that break the statistical biases of AI-generated design.
Enforces true randomization for layout variance, AIDA page structure, wide editorial typography, gapless bento grids, and strict GSAP ScrollTriggers. It exists to counter what LLMs do by default — six-line wrapped headings, cheap meta-labels, repeated left/right layouts.
design-taste-frontend
Anti-slop frontend review — an audit-first pass over UI that was already built, before it ships.
Runs after implementation, not during mocking: reads the brief, infers intent, then checks the built UI against taste rules contextually — nothing fires automatically. Strict pre-flight check before shipping.
image-to-code
Turns a single design reference image into a real, implementation-friendly frontend component.
Works image-first: reads the reference as an art director — hierarchy, spacing, tokens — then produces the component code against it as a fidelity target. Best with large section-specific images, not full-page layouts.
mock-prototype
Prototypes an idea as three HTML design directions, published as links you can open and pick from — no production code.
Runs the huashu-design pipeline: drafts three directions as HTML, hosts each on an unlisted artifact link, waits for your pick, then polishes the chosen direction and hands off a spec to the build phase.
web-design-guidelines
Reviews UI code against 105 concrete web interface rules — accessibility, focus, forms, animation, layout, content.
A heuristic engine where every rule is checkable in code, not by feel — vendored from vercel-labs/web-interface-guidelines (MIT) so it works offline. Used during review passes alongside visual screenshots.
reference
Depth on demand — routing, graph theory, and prompt craft.
figma-orientation
Router for Figma work — maps your intent to the right official Figma skill or MCP call before you guess wrong.
Loads first on any Figma task, reads what you actually want (implement a design, extract tokens, generate code), then routes to the correct skill or direct tool call. Prevents the classic failure of calling a tool without its required companion skill.
graph-engineering
Teaches graph engineering — knowledge graphs (what agents remember) and task graphs (how agents orchestrate) — in teaching mode with worked examples.
Covers ontology design, entity and relation extraction, GraphRAG serving, plus the orchestration half: parallel fan-out, verifier separation, and human gates. Ask it to build a knowledge graph or to learn the discipline, and it explains each stage while producing visual diagram artifacts.
prompt-master
Turns a rough prompt idea into one production-ready prompt, optimized for the specific AI tool you name.
Extracts the real intent, identifies the target tool — an LLM, Cursor, Midjourney, a coding agent — and outputs a single prompt with zero wasted tokens, locking identity and output format up front. Only activates when you explicitly ask for prompt work.