How to use the CLI-Anything Harness Builder agent skill

What is an agent skill and what does this one do?

An agent skill is a folder with a SKILL.md file that tells an AI agent when to use it and how to work. This skill is designed for Codex. Its description says to use it when you want Codex to build, refine, test, validate or list CLI-Anything harnesses for GUI applications or source repositories.

The skill has five modes: Build, Refine, Test, Validate and List. Build creates a full harness. Refine adds high-impact capabilities to an existing one without removing existing commands unless you ask for a breaking change. Test runs the harness tests against the real backend. Validate checks the harness against a checklist. List discovers installed and generated CLI-Anything tools.

• Generated harnesses are Python packages with one-shot Click subcommands and a default REPL mode.
• They support --json output, session state with undo/redo where the target supports it, and --dry-run.
• The skill prefers the real software backend over reimplementation.
• It keeps the generated Python harness format unchanged.

Install the skill from the ZIP

Download the ZIP from this page and unzip it into your Codex skills directory. The skill's installer defaults to ~/.codex/skills, so that is the usual location. The skill's files are agents/openai.yaml, scripts/install.sh, SKILL.md and tests/test_install.sh.

The installer, scripts/install.sh, copies canonical resources from a full CLI-Anything repository checkout into the skill, so a normal Codex installation is self-contained. You need that checkout available when you run the installer. Bash is required to run install.sh.

• Default location: ~/.codex/skills
• Requires a full CLI-Anything repository checkout for the installer to copy resources
• Runs on Bash

Trigger the skill with natural requests

You do not need a special command. Describe the outcome you want and name the software. The skill accepts either a local source path such as ./gimp or /path/to/software, or a GitHub repository URL. If you give a URL, the software name is derived from the directory name after cloning.

Name the mode in your request when you know which one you need. Phrases like build, refine, test, validate or list map to the five modes.

• Build a CLI harness for ./gimp
• Refine the existing harness for /path/to/software and add batch export
• Run the tests for the gimp harness against the real backend
• Validate the harness in ./blender against the checklist
• List the CLI-Anything tools installed on this machine as JSON

Understand what the skill reads first

Before it writes code, the skill is meant to read the full methodology. It reads references/HARNESS.md, then the matching mode specification under references/commands/, then any guides under references/guides/ that apply. If those files are missing, it looks for the plugin files in a CLI-Anything checkout, then clones or downloads https://github.com/HKUDS/CLI-Anything. Only when every full-methodology source is unavailable does it fall back to condensed rules.

This matters for you: if the methodology cannot be found, the output follows a shorter rule set. Check which source was used in the progress report.

Review the output and run the checks

A Build run produces a harness folder with an agent-harness directory, setup.py, and a cli_anything/<software>/ package containing the CLI file, core/, utils/ and tests/. The skill expects tests to be written in TEST.md before test code, with test_core.py for unit coverage and test_full_e2e.py for real-file workflows. It also asks that rendered or exported output be verified programmatically, not only by exit code.

Ask for a final report that lists the target software and source path, files added or changed, validation commands run and open risks or backend limitations.

• Confirm test_core.py and test_full_e2e.py exist and run against the real backend
• Ask for the Validate mode checklist before release
• Check the packaging: a cli_anything namespace package and a cli-anything-<software> console script

Safety notes: scripts and network access

This skill runs scripts and uses the network. install.sh is a Bash script, and the installer may clone or download the CLI-Anything repository from GitHub. Generated harnesses run commands against real backends and may modify files, so run them on copies or in version control where you can review changes.

Do not use generated harnesses on production software without review. Read the generated code before you run it, and read the install script before you execute it.

• Read scripts/install.sh before running it.
• Work on a copy of your data or a repository you can revert.
• Review generated harnesses before running them on production software.