Cli creator

Skill aizech/bernhard-zechmann-skills/skills/engineering/cli-creator

Modular AI agent skills used across engineering, writing, and research workflows. Model‑agnostic, composable, and production‑tested. Includes engineering, productivity, personal, and experimental skill packs.

Install

npx -y skills add aizech/bernhard-zechmann-skills --skill cli-creator

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Build a composable command-line tool from API docs, OpenAPI specs, SDKs, curl examples, or existing scripts. Use when the user wants a durable CLI that future agents can run from any directory.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.8 KB, as published. Nobody here has run it

CLI Creator

Build a durable CLI that agents can run from any working directory.

Start

Define:

Source: API docs, OpenAPI JSON, SDK docs, curl examples, web app, or existing script.
Jobs: literal reads/writes such as list drafts, download logs, search messages, upload media.
Install name: short binary name like ci-logs, slack-cli, sentry-cli.

Check if the name already exists:

command -v <tool-name> || true

Choose the runtime

Inspect the machine:

command -v cargo rustc node pnpm npm python3 uv || true

Choose the least surprising option:

Rust for durable, fast binaries with strong JSON handling.
TypeScript/Node when the SDK or browser tooling is the reason the CLI is better.
Python for data, notebooks, SQLite, or existing Python-heavy tooling.

Command surface

Sketch before coding. Every CLI should have:

--help showing every major capability.
--json doctor verifying config, auth, version, and endpoint reachability.
init for local config when env-only auth is painful.
Discovery commands for top-level containers.
Resolve commands turning names, URLs, or slugs into stable IDs.
Read commands with bounded --limit or clear pagination.
Write commands that are narrow, named actions with --dry-run or preview when possible.
A raw escape hatch named honestly, e.g. request or api.

Auth and config

Precedence:

Environment variable with the service's standard name, e.g. GITHUB_TOKEN.
User config under ~/.<tool-name>/config.toml.
--api-key flag only for explicit one-off tests.

Never print full tokens. doctor --json reports the auth source category and missing setup.

Build workflow

Read the source to inventory resources, auth, pagination, IDs, file flows, rate limits, and dangerous writes.
Sketch the command list in chat.
Scaffold the CLI with a README.
Implement doctor, discovery, resolve, read, and a narrow dry-run write path.
Install the CLI on PATH.
Smoke test from another repo or /tmp.
Run format, typecheck, build, and unit tests.

Defaults

Rust: clap, reqwest, serde, toml, anyhow. Add make install-local to ~/.local/bin.

TypeScript/Node: commander or cac, zod only where needed, package.json bin entry.

Python: argparse or typer, httpx/requests, pyproject.toml console script.

Companion skill

After the CLI works, create a small skill that explains how to verify it, configure auth, discover IDs, run safe reads, and use the write path.