How to use the Repomix Codebase Packer agent skill

What the repomix skill makes the agent do

Repomix is a command-line tool that packs a code project into one file. The skill tells the agent how to run it, how to read the result, and how to report back. The agent packs the repository, reads the command output for file count, character count and estimated tokens, searches the packed file with grep-style patterns, and summarises what it finds.

Repomix respects .gitignore and runs built-in security checks, according to the skill source. The skill also says the output format defaults to XML, with markdown, plain and json available.

• Packs a local directory or a remote GitHub repository (owner/repo)
• Reports files processed, total characters, estimated tokens and the output file location
• Searches the packed file for exports, auth code, API routes and database models

Install the repomix skill

Download the skill ZIP from this page. Then choose one of the following methods.

For Claude Code, unzip it into your personal skills folder ~/.claude/skills/, or into a project's .claude/skills/ folder if you want it only for that project.

In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

Requirements before you start

The skill lists Node.js with npx available as its requirement. Repomix itself is an npm package that is installed on demand through npx repomix@latest, so you do not install it separately in advance. Remote GitHub repositories also need internet access.

• Node.js with npx available
• Internet access for remote GitHub repositories
• Repomix npm package, fetched on demand

Trigger the skill with natural requests

You do not need special syntax. Ask in plain language and the agent will pick the skill when the request matches its description. The skill is written to respond to requests to explore repositories, analyse code structure, find patterns, check token counts, or prepare codebase context for AI analysis.

Name a repository in owner/repo form for remote analysis, or point at a folder for local analysis.

• "Analyze this repo" or "Explore this codebase"
• "What's the structure of facebook/react?"
• "Find all authentication-related code"
• "How many tokens is this project?"
• "Pack this repo for AI analysis"
• "Show me the main components of vercel/next.js"

Pack a repository and read the results

For a remote repository, the skill runs Repomix against the owner/repo and writes the output to a temporary directory so your working folder stays clean. For a local folder, it packs the directory you name. Large repositories can use compression, which the skill says reduces token usage by about 70%. You can also limit the files with an include pattern, for example only TypeScript and Markdown files.

The agent then searches the output file for patterns such as exports, routes or models, and reports metrics, structure, key findings and suggested next steps.

• --remote <owner/repo> for GitHub repositories
• --compress to use tree-sitter compression on large repos
• --include to focus on specific file types or folders
• --remote-branch to choose a branch, tag or commit
• --style to pick xml (default), markdown, plain or json

Safety notes: scripts and network

This skill runs scripts and uses the network. Remote mode fetches from GitHub, and packing creates a copy of your code that the agent then reads. If the code is private or sensitive and should not leave your machine, use a local-only workflow and avoid the remote option.

Repomix automatically excludes files such as API keys, credentials and .env files through built-in checks, and the skill tells the agent to trust those defaults. The skill does not guarantee that every secret is removed, so do not rely on it as your only protection for credentials.

• Remote mode sends requests to GitHub
• Packing copies your code into an output file the agent reads
• Built-in checks may miss some credentials; review sensitive projects yourself

Tips for better results

Start with a narrow question, such as where authentication lives, rather than asking for a full review of a huge repository. The skill recommends searching the output before reading it in full, and narrowing scope with --include when output is too large.

If a remote fetch fails, check the repository name and your connection, and consider cloning the repository locally instead, as the skill suggests.

• Use --compress for repositories over about 100k lines
• Use --include to narrow scope when output is too large
• Ask for a specific pattern, such as routes or models, rather than everything