How to use the CLI Developer agent skill (cli-developer)

What is an agent skill and what does the CLI Developer skill do?

An agent skill is a folder with a SKILL.md file and optional reference files. When installed, the agent reads these instructions when a task matches the skill's description, so it follows a consistent set of conventions instead of relying only on general knowledge.

This skill covers designing subcommands and flags, parsing arguments, adding interactive prompts, progress bars and spinners, and generating bash, zsh or fish completion scripts. It works with commander (Node.js), click, typer and argparse (Python), and cobra, viper and bubbletea (Go).

• Command hierarchy and flag conventions, including short and long forms and required versus optional arguments
• Configuration layers, with command-line flags as the highest priority
• Checklists such as keeping startup under 50ms, sending errors to stderr, and disabling colour when output is not a terminal
• Handling Ctrl+C (SIGINT) gracefully and supporting both interactive and non-interactive modes

How do you install the CLI Developer skill?

Download the ZIP for cli-developer from our page and unzip it. For a personal setup, place the folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.

You can also upload the skill in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the ZIP there.

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

How do you trigger the skill with natural requests?

You do not need a special command. The skill's description tells the agent when to use it, so ask for CLI work in plain language. Mentioning the framework you want helps, because the skill names commander, click, typer and cobra.

The skill is most useful when your request is about the command-line side of a program: arguments, subcommands, prompts, progress output or completions.

• "Add a config get, set and list subcommand to this commander tool."
• "Parse --input and --format flags with click and validate them early."
• "Generate zsh and fish completion scripts for our typer CLI."
• "Add a spinner while the download runs, but turn it off when output is piped."

What does a typical quick-start look like?

The skill includes a commander example that defines a command with a required argument and a boolean option. The agent is expected to follow this pattern, check the result with the help and version flags, and then add completions and error messages.

For Python and Go the skill points to separate reference files, python-cli.md and go-cli.md, for their quick-start examples.

What does the skill need, and what does it run?

The skill needs a Node.js, Python or Go toolchain, depending on the CLI you are building. The frameworks it names (commander, click, typer, cobra) are installed by your project, not by the skill itself.

The skill does not run scripts and does not use the network. It is a set of instructions and reference documents. Any commands the agent executes while building your project are the normal ones your project and environment allow, so review them as you would any code change.

• Toolchain: Node.js, Python or Go, matching the CLI you build
• Frameworks: installed by your project
• Scripts: none bundled with the skill
• Network: not used by the skill

What are the limits and safety notes?

The skill recommends testing on Windows, macOS and Linux, but it does not run those tests itself. You are responsible for running cross-platform checks and benchmarking startup time.

The skill is only useful when you are writing a CLI program. It gives reference guidance for Node.js, Python and Go; in other languages the agent will fall back on general knowledge.

• Run your own tests on each target operating system
• Check the skill's reference guidance against your framework's current documentation
• Review code the agent writes before committing it, especially around file paths and config

What tips help you get better results?

Describe the command structure you want before asking for code. The skill asks the agent to list commands and their expected help output first, which is easiest to check when you supply it.

• State the language and framework up front, for example "Go with cobra"
• Ask for non-interactive fallbacks if the tool will run in CI/CD
• Ask for completions as part of the initial build rather than later
• Keep flag names stable; the skill treats renames as breaking changes