agent-skills

Reusable skills for AI agents across multiple tools: Claude Code, Cursor CLI, Python, OpenCode, Codex.

---

Installation

npx skills@latest add motionharvest/agent-skills

This installs all available skills to your configured tools.

---

UX & Design Skills

A complete end-to-end system for website research, strategy, design, and implementation. All skills coordinate through a shared ux.md file — one source of truth for all decisions, research, personas, architecture, and design rules.

Start here: Run /perfect-design with a screenshot or vague goal. It diagnoses what's needed and routes to the right tools.

The Pipeline

Entry Point

• perfect-design — General entry point. Accepts messy input (screenshots, URLs, descriptions, half-formed ideas). Diagnoses gaps, maintains ux.md, routes to specialized skills. No UX jargon required.

Research Phase

• audience-research — Find real user pain points, barriers to entry, and language. Searches Reddit, forums, reviews, industry communities (not just tech). Grounds personas in evidence, not assumptions.

Strategy Phase

• persona-archetypes — Build psychologically-grounded personas using 50+ frameworks (Big Five, Enneagram, Jung, VALS, DISC, etc.). Map how they think, decide, and what messaging resonates.
• reference-site-analysis — Analyze 3–5 high-signal sites (validated by user reviews, market data, engagement). Extract UX patterns that actually work with this audience.

Architecture Phase

• ux-methodology-process — Determine page structure and flow using psychology-based laws (Peak-End Rule, Mental Model, Miller's Law, Hick's Law, etc.). Answers: What sections exist and in what order?

Design Phase

• ux-methodology-design — Optimize visual hierarchy, interactions, and presentation using design laws (Fitts's Law, Von Restorff, Doherty Threshold, etc.). Answers: How should information be presented?

Identity Phase

• visual-identity — Lock the brand language before writing code. Defines personality keywords, explores typography pairings, expands the color system to a full token set, and names 4 signature motion vocabulary moves. Outputs design-system.md ready for the build phase. Award-winning studios run this before implementation.

Build Phase

• motion-web-design — Build the site. Takes a build prompt (from audience-site-brief) and design-system.md (from visual-identity) and produces a coded, running landing page using Vite + GSAP + Lenis. Implements choreographed scroll animations, micro-interactions, and polished visual design. Sources imagery from Unsplash and documents video slots for Replicate. Targets Awwwards-quality execution.

Optimization Phase

• ab-testing — Plan conversion experiments before and after launch. Produces a prioritized test backlog using the PIE framework (Potential, Importance, Ease), hypothesis templates, and persona-calibrated CTA variants. Enforces the right testing order: headlines first, button color last.

Full Orchestration

• audience-site-brief — Complete pipeline for users who want guided end-to-end flow. Coordinates all phases and outputs a build prompt for the identity and build phases.

How It Works

1. Start: Run /perfect-design with a goal or screenshot
2. Diagnose: Skill asks 2–3 clarifying questions, checks ux.md for prior work
3. Route: Hands off to the appropriate skill in the pipeline
4. Coordinate: Each skill reads and updates ux.md
5. Continue: Come back anytime to check progress or move to the next phase

Example Workflows

Quick Improvement

→ Run /perfect-design with screenshot
→ It diagnoses the problem
→ Routes to /ux-methodology-design (if visual) or /audience-research (if strategy)
→ Updates ux.md with findings

New Site From Scratch

→ Run /audience-site-brief with project goal
→ Orchestrates: /audience-research → /persona-archetypes → /reference-site-analysis
→ Then: /ux-methodology-process → /ux-methodology-design
→ Then: /visual-identity → design-system.md
→ Then: /motion-web-design → coded, running site
→ Then: /ab-testing → prioritized test backlog for launch

Copy & Messaging Only

→ Run /audience-research to find pain points
→ Then /persona-archetypes to build personas
→ Use findings to write better copy in /ux-methodology-design

See SKILLS.md for detailed skill descriptions, dependencies, and tool support.

The ux.md File

All UX skills coordinate through a shared ux.md file in your project. This file:

• Tracks all research — pain points, language, objections from real users
• Documents personas — psychology, decision style, trust signals
• Records decisions — what sections exist, why, in what order
• Captures design rules — visual hierarchy, emphasis, interaction principles
• Provides visibility — one source of truth for the entire design process

Each skill reads ux.md first to avoid duplicate work and understand context. Each skill updates it with new findings.

---

Dev Tools

inspector-overlay

Add a DOM inspector overlay to React or Next.js apps. Installs a development-only component that lets you hold Alt to inspect elements, Alt+click to copy single-element payloads for coding agents, and Alt+Shift+click to build multi-element selections. Payloads include semantic hooks (data-testid, data-file, data-component) that make AI-assisted edits dramatically more precise.

Use when you want to point a coding agent at a specific element without describing it in words.

---

CAS Skills

Conscious Agentic System infrastructure for reliable, auditable agent behavior. These skills enforce a structured approach to agent reasoning and learning.

cas-agent-setup

Set up CAS persistence for a new or existing agent. Creates the required files and directories (STATE.json, append-only telemetry, raw output store, optional archive layout) and wires schemas and specs. Ships bundled helpers: STATE template, all JSON schemas, consolidation spec, and acceptance test definitions.

Use when bootstrapping an agent or aligning an existing agent with the CAS file contract.

cas-operational-loop

Run the 8-phase CAS loop for every consequential action. Phases: State Ingestion → World-Model Update → Prediction → Action → Evaluation → Learning → Commit. Covers prediction/observation/learning record formats, raw output capture, pending citations, run_nonce scoping, and error classification.

Use when operating as a CAS agent — this skill defines the how, while cas-agent-setup defines the what files.

Why CAS?

Most agents fail because they:

• Act without grounding in current context
• Forget why they made a decision
• Don't learn from mistakes, so failures repeat

CAS solves this by making reasoning and learning operational. It enforces:

• Traceability — every action has a prediction + evaluation
• Continuity — session-to-session improvement without re-explaining context
• Error learning — failures are classified and followed by explicit learning
• Grounded execution — actions tied to observed state, not vague intent