How to Use Claude Skills: Practical Setup Guide (2026)
From Prompts to Skills
Standardize Workflows: Ensure every code review or commit message follows the exact same format.
Execute Code: Skills can contain Python/Bash scripts that run locally (e.g., to lint code or query a database).
Share Expertise: Package your best prompts and scripts into a skill folder with a SKILL.md file and share it with your team.
Level 1: Installing Available Skills
- Option A: Using Claude Code (CLI). Open your terminal and use the currently supported plugin flow shown in Anthropic documentation or the repository README.
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills
- Option B: Using a platform-specific visual manager. The steps below describe a possible Verdent-style workflow and should be treated as product-specific guidance rather than Anthropic documentation.
Open your platform interface.
Go to the area where installed tools or skills are managed.
Review the available packages, verify the source, and install the one you need.
On platforms that support multi-agent workflows, you may choose to separate skills by role, such as frontend review versus backend review.
Level 2: Authoring Your First Skill
Create the File Structure
my-skills/ └── strict-reviewer/ └── SKILL.md
Write the Frontmatter (YAML) This tells Claude when to use the skill.
---
name: strict-review
description: Reviews code for security vulnerabilities and performance issues. Use when the user asks for a PR review or to check code.
---
# Code Review Guidelines
Write the Instructions (Markdown) This tells Claude how to perform the task.
# Code Review Guidelines You are a strict code reviewer. For every file provided: 1. Check for **OWASP Top 10** vulnerabilities. 2. Identify O(n^2) or worse algorithmic complexity. 3. Ensure variable naming follows snake_case for Python. Output format: - 🔴 [Critical]: <Issue> - 🟡 [Warning]: <Issue>
Make the skill available by placing it in Claude’s skills directory, such as ~/.claude/skills/ for personal use or .claude/skills/ for project-specific use.
Level 3: Adding Executable Scripts
Add a Script Create run_lint.py in your skill folder.
# strict-reviewer/scripts/run_lint.pyimport sys # ... logic to run pylint and capture output ... print(lint_report)
Update SKILL.md Tell Claude to run this script.
... Before analyzing the code manually, run the linter script to get objective metrics: `python scripts/run_lint.py --target {target_file}` Use the output of this script to inform your review. ...
- Now, when you invoke the skill, Claude will autonomously execute the Python script on your machine, read the output, and generate a review based on real data.
Claude Skills Cookbook
Recipe 1: The Commit Message Standardizer
- Goal: Force all git commits to follow "Conventional Commits".
- Frontmatter: name: commit-msg, description: Generates git commit messages.
- Instruction: "Always format as <type>(<scope>): <subject>. Types allowed: feat, fix, docs..."
Recipe 2: The Data Analyst
- Goal: Visualize CSV files.
- Script: plot_data.py (Uses Matplotlib to generate a .png).
- Instruction: "When given a CSV, run plot_data.py to generate a histogram, then describe the trends seen in the image."
Recipe 3: The Documentation Updater
- Goal: Keep READMEs in sync.
- Instruction: "Read src/ files. Compare with README.md. If functions are missing documentation, generate a diff to add them."
Level 4: Orchestration in a Platform Layer
The "Parallel Review" Workflow:
Select files: choose a batch of modified files in your workspace.
Select a skill: choose your custom strict-review skill.
Action: Click "Parallel Run".
Result: the platform may spin up isolated task runners or agents, each applying the same skill to a different file.
Merge the results back into one review summary.
Frequently Asked Questions
Where do I save my custom skills?
~/.claude/skills for personal skills and .claude/skills for project-specific skills. Platform-specific management layers may provide additional ways to organize those paths.Can a skill call another skill?
How do I debug a skill that isn't working?
SKILL.md parses correctly, and confirm that any referenced script runs successfully in your environment. Some platforms may also provide additional logs.