pstack's own behaviour belongs upstream at cursor/plugins. This repository rewrites and republishes it, so a skill that gives bad advice in Cursor gives the same advice here.

Report to this repository what the port itself introduces:

• A generated skill that points at a path or a tool that does not exist. An agent will act on that instruction anyway, which is the whole reason port/verify.mjs fails the build over it.
• A rewrite in port/rules.mjs that changed what an instruction means, rather than which product it names.
• Anything in port/ or .github/ that would let text fetched from upstream reach this repository's write token.

Open a normal issue for the first two. For the third, use private vulnerability reporting on the Security tab, so a fix ships before the details do.

Nothing from upstream. port/sync.mjs fetches the pstack/ subdirectory with a blobless sparse checkout, reads the files, rewrites text, and writes the result. It never executes them, and the workflow installs no dependencies.

pstack ships shell and TypeScript scripts that its skills ask an agent to run, four of them executable. Read them first if that matters where you work. UPSTREAM.json names the exact upstream commit every file came from, so you can diff the published tree against cursor/plugins yourself.