hf-cli skill pipes an installer into a shell; heads-up on an audit publishing September 8

One skill instructs Claude to install a CLI by piping curl into a shell:

• skills/hf-cli/SKILL.md:6 runs curl -LsSf https://hf.co/cli/install.sh | bash -s

When I read this pack at commit 87f9ee5 on July 31, skills/hf-cli/SKILL.md:227 also fetched an install script from an unpinned main branch (raw.githubusercontent.com/huggingface/hf-mount/main/install.sh | sh). That line is gone at current main; the text now says brew install hf-mount, which is the change I would have suggested, so there is nothing to ask for there. The writeup will note the fix, with the date.

I want to be straightforward about the remaining one: piping an installer into a shell is an ordinary way to ship a CLI, and I wouldn't call it reckless. It's flagged because fetch-and-execute inside an instruction file is something my rubric measures. An instruction file is loaded into an agent's context, so a line telling the agent to run an installer is a slightly different thing from a line in a README telling a human to.

Context, so this isn't arriving without explanation: I audited twelve Claude Code plugins and this pack is one of them. It comes out well: no hooks, source pinned to a commit, nothing hidden, clean prompts. It's graded B, and these lines are the only reason. I know this file is synced from huggingface_hub; I'm filing here because this is the repository the marketplace pins, and I'm happy to move it.

The writeup publishes Tuesday, September 8, at plugrade.dev, and the rubric plus what it can't see is at plugrade.dev/methodology. If you think the grade is wrong, I'd genuinely like to print why. Replies go up verbatim next to the finding, unedited, and "no comment" is a fine answer.