CLI Anything

CLI-Anything for Codex

Use this skill when the user wants Codex to act like the CLI-Anything builder.

Read the Full Methodology First

Before implementation, use the full methodology source of truth:

1. Read references/HARNESS.md relative to this skill directory.
2. Read the matching mode specification under references/commands/.
3. Read referenced files under references/guides/ only when they apply to the target.
4. If the installed resources are unavailable and this skill is being used from a CLI-Anything repository checkout, read ../cli-anything-plugin/HARNESS.md and the resources around it.
5. If neither local source is available, clone or download https://github.com/HKUDS/CLI-Anything and use cli-anything-plugin/HARNESS.md.
6. Only if all full-methodology sources are unavailable, follow the condensed rules below.

The installer vendors the canonical resources from cli-anything-plugin/ into this skill so a normal Codex installation is self-contained.

Resource Map

When reading vendored documents, apply these path remappings instead of resolving plugin paths against the current working directory:

Inputs

Accept either:

• A local source path such as ./gimp or /path/to/software
• A GitHub repository URL

Derive the software name from the local directory name after cloning if needed.

Modes

Build

Read references/commands/cli-anything.md. Build a complete harness through source analysis, architecture design, implementation, test planning, test implementation, test documentation, CLI-specific SKILL.md generation, and local installation.

Refine

Read references/commands/refine.md. Inventory current commands and tests, compare them against the target software, then incrementally add high-impact capabilities without removing existing commands unless the user requests a breaking change.

Test

Read references/commands/test.md. Run the harness tests against the real backend, verify installed-command subprocess coverage, and update TEST.md only with passing results.

Validate

Read references/commands/validate.md. Validate the harness against the complete directory, implementation, test, documentation, packaging, and code-quality checklist.

List

Read references/commands/list.md. Discover installed and generated CLI-Anything tools and support human-readable or JSON output as requested.

Condensed Fallback Rules

Use these only when the full methodology cannot be retrieved.

Harness Structure

Produce:

<software>/
└── agent-harness/
    ├── <SOFTWARE>.md
    ├── setup.py
    └── cli_anything/
        └── <software>/
            ├── README.md
            ├── __init__.py
            ├── __main__.py
            ├── <software>_cli.py
            ├── core/
            ├── utils/
            └── tests/

Required Behavior

• Prefer the real software backend over reimplementation.
• Provide one-shot Click subcommands and default REPL mode.
• Support --json machine-readable output.
• Add session state with undo/redo where the target supports it.
• Auto-save one-shot session mutations and support --dry-run.
• Use locked session-file writes to avoid concurrent JSON corruption.
• Copy and use the unified ReplSkin.
• Add truthful preview commands for software with meaningful visual or inspection state.
• Generate both canonical and packaged CLI-specific SKILL.md files.

Testing

• Write TEST.md before test code.
• Keep test_core.py for unit coverage.
• Keep test_full_e2e.py for real-file workflows and real backend validation.
• Verify rendered/exported output programmatically, not only process exit codes.
• Test the installed cli-anything-<software> command via _resolve_cli().
• Run release validation with CLI_ANYTHING_FORCE_INSTALLED=1.

Packaging

• Use find_namespace_packages(include=["cli_anything.*"]).
• Keep cli_anything/ as a namespace package without a top-level __init__.py.
• Expose cli-anything-<software> through console_scripts.
• Include the packaged CLI-specific skills/SKILL.md.

Output Expectations

When reporting progress or final results, include:

• target software and source path
• files added or changed
• validation commands run
• open risks or backend limitations