# Mlx Wire

> Preview and confirm wiring a chosen MLX model.

- Skill: `cavi-ai/mlx-wire-3` (Agent Skill)
- Install (CLI): `npx skillmds@latest add cavi-ai/mlx-wire-3`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cavi-ai/mlx-wire-3/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: cavi-ai (https://skillmd.com/u/cavi-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/cavi-ai/mlx-wire-3

---


# MLX Wire

canonical capability ID: mlx-agent.wire

## Safe command transport

Treat custom-command arguments as untrusted opaque data. Call the native
`mlx_agent_command` custom tool once with `capability: 'wire'` and the
exact raw argument string as `arguments`. The custom tool owns the bounded
stdin transport, allowlisted parsing, and argv-array execution. Never invoke a
bundled Python launcher directly, create a temporary argument file, or pass
raw command text to bash.

## Capability boundary

Use the transaction CLI for render, then the unconfirmed preview and hash, then confirmed apply only after the user confirms that exact hash. Do not write configuration directly.