[liogi]

0.8.4

Nothing to reinstall — the shell wrapper is unchanged. gwt sync-env copies
fewer files than it did, on purpose; the two entries below say which and why.

Fixed

• gwt sync-env overwrote committed .env templates. The scan skipped
  .env.tpl, .env.example and their three siblings by exact name, so every
  prefixed form went through — .env.development.tpl, .env.production.tpl,
  .env.development.example. Those files are usually tracked by git, so the main
  worktree's copy landed on top of a branch that had edited one, and git status
  showed a change nobody made. Run git add -A after that and you commit the
  main worktree's template into your branch. One monorepo had 44 such files.

  The check is on the suffix now, so .env.<anything>.tpl is skipped whatever
  environment it names. .env.test.local is still copied — it is a local
  override, not a committed fixture.

• And a suffix list is only a guess, so git decides. A repository that
  versions .env.defaults — no template suffix anywhere — had the same bug in a
  different shape. sync-env now leaves alone any file git tracks in the
  destination, whatever it is called, and says how many it left. The suffix rules
  stay on top of that, because they carry an intent an index cannot: an untracked
  .env.example is still an example, and the guard yields nothing when git fails
  or when a repository has committed nothing yet.

  Measured on the monorepo that surfaced this: of 68 .env* files, the 40 git
  tracks are now left alone and the 28 it does not are still copied — nothing on
  the wrong side of either rule.

Internal

• actions/checkout and actions/setup-node go from v4 to v7, which ends the
  Node 20 deprecation warning on the runners. setup-node v7 also drops the
  dummy NODE_AUTH_TOKEN export that made registry-url return E401 against
  npm's trusted publisher, so that workaround is no longer load-bearing — it
  stays anyway, since the publish path works.

• The tracked-file lookup is one git ls-files scoped by pathspec. A bare
  listing measures 1.09 MB on a large monorepo, past execFileSync's default
  buffer, against 1.7 KB scoped.

• Four worktree colours collide 8.8% of the time, measured over 20,000 draws.
  The code said eight percent in one place and one in fourteen in another, and
  0.8.3 below says seven; the first was closest. The comments and the test now
  carry the measured figure. No colour changes.

---

After merging

git tag v0.8.4 && git push --tags

CI stages the package; approve it on npmjs.com with 2FA. 165 tests.