fix(install): clear blockers before brew cask install and surface daemon errors

[AnnatarHe]

Problem

On macOS, curl -sSL https://shelltime.xyz/i | bash could fail the Homebrew install and quietly fall back to the manual install, leaving the daemon service on the previous release:

• An unmanaged regular file at $(brew --prefix)/bin/shelltime-daemon (written by older shelltime update runs) makes the cask's binary link fail with "It seems there is already a Binary at …".
• Users from before the formula→cask migration still have a shelltime formula keg. tap_migrations.json maps shelltime to the same tap and name, which Homebrew treats as a no-op, so these users are never migrated.
• shelltime daemon reinstall output went to /dev/null, which hid failures.

Changes (macOS Homebrew branch of install.bash)

• Run brew update --quiet so a reinstall sees the newest cask despite brew's auto-update throttle.
• Remove regular (non-symlink) shelltime / shelltime-daemon files from brew's bin dir before installing.
• If $(brew --cellar)/shelltime exists, brew uninstall --formula shelltime.
• Install with brew install --cask shelltime/tap/shelltime.
• Capture shelltime daemon reinstall output and print it to stderr when the command fails.

Companion CLI fix: shelltime/cli fix/daemon-binary-resolution. It makes a curl-installed CLI always use its own daemon and stops shelltime update from writing into the Homebrew prefix.

Testing

• bash -n install.bash passes.
• Not yet run end to end: needs a macOS machine with Homebrew. That run should start from a stray /opt/homebrew/bin/shelltime-daemon regular file and the old formula keg installed, and should end with shelltime daemon status reporting the latest version.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[claude]

Review

Clear, well-commented fix that targets the three failure modes described. bash -n passing is good, but the real behaviour is untested (see Tests).

Looks good

• Output of shelltime daemon reinstall is captured and sent to stderr only on failure, so errors are no longer hidden.
• The -f and ! -L guard means Homebrew-managed symlinks are never touched.
• The script has no set -e, so a failing brew update doesn't abort the install.

Suggestions

1. Destructive removal happens before the install is known to succeed. If brew install --cask fails, the removed regular file may have been the user's only working binary. The fallback reinstalls to ~/.shelltime/bin, so the risk is low. Moving it to *.bak instead of rm -f would be safer and match the existing rename step. rm failures (for example permissions) are also unchecked, so "Removed..." can print when it wasn't. Consider rm -f ... && echo ....
2. The result of brew uninstall --formula shelltime is ignored. If it fails, the cask install will fail with a confusing conflict, so a warning would help.
3. brew list --formula shelltime >/dev/null 2>&1 is a slightly more idiomatic check than testing the Cellar path.
4. brew update --quiet adds latency to every install. A short "Updating Homebrew..." message would keep the pause from looking like a hang.
5. bin and daemon_output are globals. This is harmless, but bin is a generic name.

Tests

• CI runners lack the old-formula and stray-file preconditions, so none of the new branches are covered. Please run the manual macOS scenario from the description before merging. A macOS CI step that creates a dummy regular file at $(brew --prefix)/bin/shelltime-daemon and then runs the installer would give lasting coverage.
• The companion CLI PR (fix/daemon-binary-resolution) should be released together with this one, since the daemon version outcome depends on both.

No security concerns beyond the rm scope above: paths come from brew --prefix and fixed names.

🤖 Generated with Claude Code