fix(progress): name the running operation in watchdog status lines

[max-sixty]

The watchdog status line named the awaited artifact — ↳ Waiting for the commit message (4s) — which reads like waiting on user input. It now names the running operation:

• LLM shell-outs (commit, squash, wt config show connectivity test): Waiting for the commit generation command (4s), reusing the existing vocabulary ([commit.generation] command config key, "Commit generation command failed" error title). All three sites run the same configured command, so the phrase moves into watch_llm_command and its waiting_for parameter is dropped.
• Version check in wt config show --full: Waiting for the version check (4s)
• wt switch pr:N / mr:N: Waiting for the PR lookup (4s) / the MR lookup

The escalated form now reads coherently, with the status naming a command and the gutter revealing it:

↳ Waiting for the commit generation command (12s)
   sh -c 'claude -p --model=haiku'

The version-check watchdog guard now spans the whole fetch_latest_version lookup rather than just the curl block; nothing renders inside its 2s startup delay, so behavior is unchanged, and the line executes under the existing config-show tests (codecov/patch flagged it as an uncovered patch line in the network-only path).

This was written by Claude Code on behalf of max

[worktrunk-bot]

Wording change reads cleanly — the watchdog now names the running operation ("Waiting for the PR lookup", "the version check", "the commit generation command") rather than the awaited output, and the doc comments plus every test assertion are kept in sync. The watch_llm_command signature drop is reflected at all three call sites. No stale wording remains anywhere in the tree. Nothing to change in the code.

Not approving only because codecov/patch is red — and it's worth noting why, since it's a false alarm for this PR. The single uncovered line is src/commands/config/show.rs:1518, the string-only edit inside fetch_latest_version(). That line was already uncovered before this PR — it lives on the curl → api.github.com path that tests don't exercise by design — and codecov re-flags it as a new patch line purely because the literal changed. The other three touched files (progress.rs, llm.rs, switch.rs) are 100% covered.

There's no meaningful test to add here (covering it would mean mocking the version-check network call for a string swap), so this is an acceptable pre-existing gap. Per CLAUDE.md, merging past a red codecov/patch is your call.

[worktrunk-bot]

codecov/patch failed on HEAD, and this commit is the cause — it made the gap larger, not smaller. Dismissing my earlier approval accordingly.

The reword goal succeeded: moving Watchdog::start to function entry (line 1500) does get it covered by the existing config-show tests. But lifting it out of the let output = { … } block re-indented the whole curl call from 8 spaces to 4, so Codecov's line-diff now treats all 13 curl-block lines as newly added. Those lines live on the api.github.com path that tests short-circuit via WORKTRUNK_TEST_LATEST_VERSION, so they're uncovered — patch coverage for show.rs is 7.14% (1 hit / 14 lines). Before this commit the patch had exactly one uncovered line (the reworded label); now it has thirteen.

Two ways out:

• Keep the block wrapper (recommended — gets codecov/patch green). Leave the watchdog at the top where it is now, but keep let output = { Cmd::new("curl")…run()? }; as a block so the curl lines stay at their original indentation. Codecov then sees only the watchdog line in the diff (covered), and the never-tested curl lines drop out of the patch entirely since their bytes are unchanged from the base:

  let _watchdog = worktrunk::progress::Watchdog::start("the version check", None);
  // …
  let output = {
      Cmd::new("curl")
          .args([
              // …unchanged…
          ])
          .run()?
  };

• Revert this commit. That returns to the single-line gap my prior review already judged an acceptable pre-existing gap (the network path is untestable by design), which per CLAUDE.md is a red codecov/patch you can merge past at your discretion. Simpler, but loses the coverage the reword gained.

Happy to push either — say the word and I'll open a follow-up commit on this branch. Everything else in the PR (the wording change and its tests) is unaffected and reads cleanly.

codecov/patch failed — this commit re-indented the untested curl block into the patch, taking uncovered added lines from 1 to 13. See follow-up review.