Skip to content

One-click install for agentic tooling (Claude Code plugin + npx-runnable CLI) #8

Description

@oveddan

#2 makes the skill exist. This makes it installable in one step.

The current path

Per docs/agent-skill.md on agent/claude-codex-skill (#4), a new user has to:

  1. clone the repo
  2. npm install
  3. npm run build
  4. cp -R .claude/skills/mft-configurator ~/.claude/skills/
  5. cp -R .claude/skills/mft-configurator ~/.codex/skills/

Five steps and a local toolchain, before touching a knob. The README (#7) claims "you describe what you want and it gets done" — that claim is undercut by a five-step manual install.

What one-click should mean

Claude Code — a plugin, not a copied directory. Ship .claude-plugin/plugin.json with the skill bundled, plus a marketplace manifest in this repo, so installation is /plugin marketplace add oveddan/mft-api followed by /plugin install. Verify the exact manifest shape and commands against the current Claude Code plugin docs before implementing — do not copy the shape from this issue.

The CLI — runnable without a clone. npx @oveddan/mft-config export should work. That removes steps 1–3 entirely and means the skill can reference a stable command instead of node dist/cli.js.

Codex — a documented one-liner. There is no marketplace equivalent to target, so the honest goal is a single copy-pasteable command rather than pretending parity exists.

Acceptance criteria

  • A user with neither the repo nor Node experience can install and invoke the skill in Claude Code without cloning
  • npx runs the CLI against a connected Twister with no local build
  • Codex install is a single documented command
  • The skill references the published command, not node dist/cli.js
  • README install section replaced with the one-click path; the from-source path moves to a "develop" section

Depends on


Alternative to explore later

Dan, 2026-08-07: "it can also be like a command that you run or program you install or something, that Claude talks to."

Rather than a skill that shells out to a CLI per invocation, the tool could be a long-running local process the agent connects to — an MCP server, or a small daemon with a documented endpoint. Install it once, and every agent that speaks the protocol can talk to it.

Why it might be better than the skill-plus-CLI shape:

  • The Twister connection stays open instead of being rediscovered on every command.
  • One install serves Claude Code, Codex, and anything else that speaks MCP — no per-tool skill directory to copy and keep in sync.
  • Typed tool definitions instead of the agent composing shell strings, which is a narrower and safer surface for a tool that writes firmware settings over USB.
  • It matches how chromatik-mcp already works, so the two would install and behave alike.

Open questions before committing to it:

  • Does it replace the skill in Add a Claude Code and Codex compatible MFT skill #2, or sit under it?
  • Does a persistent process holding the MIDI port conflict with the DAW or the stock Twister editor also wanting it?
  • Is the plan/apply flow still legible to a human when it's tool calls instead of copy-pasteable commands?

Not scoped yet — recorded as an option, not a decision.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions