[drona23]

What & why

Gemini and Antigravity lifecycle-hook JSON stored unquoted launcher and shim paths. On macOS/Linux, a valid user-selected hive path containing a space was split by /bin/sh, so both commands exited 127 before either shim started.

Use the existing quoted nodeRun() form for these two providers on POSIX while preserving their legacy Windows command shape. The regression executes the exact generated commands with a stripped hook PATH and verifies that both shims reach HIVE_SOCK.

Closes #364

This intentionally excludes Codex, which is already handled by #356, and does not alter the separate Windows cross-shell contract tracked in #350.

Type of change

• Bug fix
• New feature
• Refactor / cleanup
• Docs
• Build / CI

Evidence

Before

After

How I tested it

• OS: macOS arm64
• node --test test/hive-hook-node.test.cjs — 11/11 pass
• npm run test:focused — 746/746 pass
• npm run typecheck — node and web pass
• npm run build — production build succeeds
• git diff --check — clean

Checklist

• Before and after evidence is attached above.
• npm run typecheck passes.
• npm run test:focused passes.
• npm run build succeeds.
• This PR is one change.
• I read the diff and removed the overlapping Codex/dependency changes.
• No UI tokens or art are changed.

[chaitanyagiri]

Reviewed by an agent on the Munder Difflin hive floor, posted from @chaitanyagiri.
First person statements below are the agent's, including the scope limits.
Replies here are read. Questions in this comment are real questions.
Base: origin/main at 956bfb4.

---

The comment rewrite is the honest part of this and worth keeping as written. The old one claimed
the hive root is space-free by construction, and the new one says the user selected path may
legitimately contain spaces. That assumption being wrong is the whole of #350 and #364, so
correcting the comment matters as much as correcting the code.

The test runs the generated command string for real and asserts both shims reach HIVE_SOCK,
which is much better than asserting the string looks right. That said, it proves the string works
under whatever run() executes it with, and the real consumer is agy's own hook runner. If agy
splits the command field on whitespace on POSIX instead of handing it to a shell, quoting breaks
it and this test would still pass. You clearly know the executor differs per platform, since
win32 keeps nodeRunUnquoted. Can you say in the PR how you established that the POSIX side goes
through a shell? If it was measured against agy rather than inferred, that is the sentence that
makes this reviewable.

Minor: nodeRun quotes the launcher and the script but joins ...args raw. Every current caller
passes a fixed event name with no spaces so it is fine today. If an arg ever carries a path, it
will not be quoted.