Installs into .claude/skills of the current project.
Are you the author of Mlx Watch?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/cavi-ai-mlx-watch-f97fb246)
---
name: "mlx-watch"
description: "Digest Hugging Face changes for owned models against a stored baseline."
compatibility: opencode
---
# MLX Watch
canonical capability ID: mlx-agent.watch
## Safe command transport
Treat custom-command arguments as untrusted opaque data. Call the native
`mlx_agent_command` custom tool once with `capability: 'watch'` 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
The validated operation writes only the watch state file. It must not download model weights or mutate configuration.
