How to use the CLI-Anything for Hermes agent skill

What an agent skill is and what this one does

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches the skill's description. This skill's description says it applies when you want Hermes Agent to build, refine, test or validate a CLI-Anything harness for a GUI application or source repository.

The skill adapts the CLI-Anything methodology to Hermes without changing the generated Python harness format. It has three modes: Build for a new harness, Refine for an existing one, and Test and Validate for checking the result.

• Build: creates an agent-harness folder with a setup.py and a cli_anything/<software>/ package
• Refine: inventories current commands and tests, then proposes gap-filling additions without removing existing commands
• Test and Validate: checks for test_core.py and test_full_e2e.py, a working entry point, JSON output and a REPL default

Install the cli-anything-hermes skill

Download the ZIP from this page and unzip it. For a personal setup, place the skill folder in ~/.claude/skills/. For a single project, place it in the project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

The skill also includes scripts/install.sh, a helper script that copies the skill into the Hermes skills folder. Read it before running it.

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

Trigger the skill with natural requests

You do not need a special command. Describe what you want in plain language and mention the source. The agent accepts either a local source path, such as ./gimp or /path/to/software, or a GitHub repository URL. It derives the software name from the directory name.

Name the mode you want if you have one in mind: a new harness (build), changes to an existing harness (refine), or checking an existing one (validate).

• Build a CLI-Anything harness for ./gimp using Hermes
• Refine the existing harness in ./inkscape and find missing features
• Validate the harness at /path/to/software for JSON output and an installable entry point
• Create a harness from https://github.com/owner/repo

What the agent produces

For a build, the skill specifies a Click-based CLI with one-shot subcommands, a REPL that starts when no subcommand is given, a --json flag for machine-readable output, and session state with undo and redo where the target software supports it.

The skill prefers wrapping the real software backend in utils/<software>_backend.py rather than reimplementing it. It also asks the agent to write TEST.md before the tests, and to verify the install with pip install -e .

• Packaging uses find_namespace_packages(include=["cli_anything.*"]) and a console_scripts entry named cli-anything-<software>
• Tests are run through the installed command via subprocess where possible
• The final report lists the source path, files changed, validation commands and open risks

What you need before starting

The skill requires Hermes Agent as the runtime. Generated harnesses are Python packages, so you need Python and pip to install them. If the agent must fetch the cli-anything-plugin methodology or a GitHub repository, it needs Git or internet access. The helper script runs under Bash.

The skill runs scripts and uses the network, so plan for both.

• Hermes Agent (runtime)
• Python and pip
• Git or internet access for GitHub sources or the cli-anything-plugin methodology
• Bash for scripts/install.sh

Safety notes before you run it

The agent runs terminal commands, installs packages and may fetch code from GitHub. Do not point it at a sensitive codebase without reviewing what it does. Read scripts/install.sh before running it, and review generated files and commands as they are produced.

The skill reads ../cli-anything-plugin/HARNESS.md first when run inside the CLI-Anything repository. If that is missing, it tries to clone or download cli-anything-plugin from the HKUDS/CLI-Anything repository. Only if both fail does it use its condensed rules.

• Run it on a copy or a non-sensitive project first
• Review install.sh and any package installs
• Check which source the agent fetched from GitHub

Tips for better harnesses

Give the agent a concrete source path or repository, because the harness is built from that source. Ask for validation after a build so the JSON output, REPL default and entry point are checked. For existing harnesses, ask for a refine pass rather than a rewrite, since the skill avoids removing commands unless you explicitly request a breaking change.

For an overview of supported harnesses and their backend patterns, the skill points to registry.json at the repository root.

• Start with Validate on an existing harness to see its gaps
• Ask for TEST.md and tests to be written and run
• Keep the harness wrapping the real backend where one exists