Claude Skills Explained: My Workflow for Creating Reusable AI Agents with Cursor and Claude Code

By: Claire Vo (ChatPRD) Source: ChatPRD — How I AI Type: article

Summary

Walkthrough of three workflows for creating Claude Skills — Anthropic’s feature for reusable, on-demand instructions. Skills are folder-based (SKILL.md + supporting files), invoked by natural language, and portable between Claude Code and the web UI.

Key Topics

What Claude Skills Are

Reusable micro-instructions callable in any conversation (unlike Projects, which maintain constant context)
A skill = a folder with SKILL.md (YAML metadata + natural language instructions) plus optional templates, examples, scripts
Defined in natural language; can reference local files and run Python scripts
Uploaded as .zip to claude.ai or used directly via Claude Code in the skill directory

Workflow 1: Claude’s Built-in Skill Creator (Learning)

Claude has a meta-skill for generating other skills
Output is very detailed: decision trees, clarifying questions, keyword triggers, templates
Downsides: over-generates files (12 when ~5 needed), download failed with file-not-found error
Value: educational — shows what a high-quality skill structure looks like

Workflow 2: Cursor + Claude Code (Recommended)

Create a claude-skills folder, open in Cursor
Build a meta-skill (create_skill) by giving Cursor the Anthropic Skills docs as context
Cursor also generated a Python validation script (validate_skill.py) — checks YAML, file refs, structure
Invoke the meta-skill via Claude Code to generate domain-specific skills (e.g., changelog-to-newsletter)
Claude Code auto-discovers skills in the working directory — no special invocation needed

Workflow 3: Uploading to Claude Web UI

Zip the skill folder → upload to claude.ai
Constraint: skill names must be lowercase, letters/numbers/hyphens only
Once uploaded, skill is immediately available in chat

Meta-Skill Pattern

A skill whose purpose is generating other skills — a reusable factory
Demonstrates composability: skills can invoke and build on other skills
The recommended starting point for any skills workflow

Key Quotes

“Unlike Claude Projects or custom GPTs, which keep a constant context within a specific chat, Skills are tools you can call on whenever you need them.”

“When you start building small, modular skills instead of just writing one-off prompts, you can make the work you get from AI more consistent and higher quality.”

Key Ideas Extracted

Meta-Skill Pattern — Build a “skill that builds skills” to bootstrap agent capabilities consistently; factory pattern for agent skill creation

Relevance to AI PM

horizontal/agents: Meta-skill pattern is a concrete example of agent skill composition and factory patterns
horizontal/context: Skills as a structured approach to context management — bundling instructions, templates, and validation together
build: The changelog-to-newsletter and demo-to-follow-up skills are practical examples of PM workflow automation