pnpm install preflight (skip cold-cache reinstalls)

When to use this: every lich up spends 30-60 seconds running pnpm install even though the lockfile hasn't changed. A before_up hook that always runs pnpm install is the obvious first attempt; the problem is it always runs.

The fix is to check whether node_modules is stale relative to the lockfile and only reinstall if so. The pattern compares pnpm-lock.yaml's mtime to node_modules/.modules.yaml's mtime (pnpm writes the latter on every install), plus a sanity check that node_modules/.pnpm exists at all.

lifecycle:
  before_up:
    - cmd: |
        set -euo pipefail
        if [ ! -d node_modules/.pnpm ] \
           || [ ! -f node_modules/.modules.yaml ] \
           || [ pnpm-lock.yaml -nt node_modules/.modules.yaml ]; then
          echo "[preinstall] lockfile changed or node_modules missing — installing"
          pnpm install --frozen-lockfile
        else
          echo "[preinstall] node_modules up to date — skipping install"
        fi

Equivalent shapes for the other package managers:

• yarn (berry): check pnpm-lock.yaml → yarn.lock; check node_modules/.modules.yaml → .yarn/install-state.gz.
• npm: check pnpm-lock.yaml → package-lock.json; check node_modules/.modules.yaml → node_modules/.package-lock.json.

This pattern may eventually become a built-in (something like runtime.preinstall: pnpm) — until then, the manual before_up hook is the recipe.

Common mistake: running pnpm install unconditionally on every lich up. It wastes 30-60s per startup and slows down agent-driven workflows where lich up/lich down happen frequently.