[JacobStephens2]

Summary

Every line thirdshift prints on stderr now starts with the local time, so a quiet terminal shows how long the Run has been on its last step:

thirdshift: 12:14:49 implement: $ cargo test 2>&1 | tail -40
thirdshift: 12:25:03 CI on 26c8b06: 2 of 5 checks still running

 progress::step(message)
-  writeln "thirdshift: {message}"
+  writeln "thirdshift: {HH:MM:SS now} {message}"

 spec_run::finish_ticket  (relays a Ticket's child stderr)
-  strip "thirdshift: ", then step("#N: {line}")
+  progress::relay("#N", line)
+    keep the child's stamp: "thirdshift: {child's time} #N: {message}"
+    (no stamp from the child → stamp it now)
+    return the message, without the stamp, for the Ticket's failure cause

• branch.rs's one direct eprintln! goes through progress::step now, so it gets stamped as well.
• In the tests, RunResult.stderr has the stamps removed by unstamped(), so the existing stderr assertions pass unchanged. The stamped text is kept as stamped_stderr.
• The README explains the stamp and adds it to the Failed spec run example.

The Run notification, the Failed run comment, the PR body and session logs don't change: none of them read the stamped stderr.

Evidence

New tests, which fail on main and pass here:

progress_lines::every_line_starts_with_the_local_time_it_was_printed
  run an issue
  each stamped stderr line == "thirdshift: HH:MM:SS …", with a time between the clock readings taken before and after the Run

spec_run::each_relayed_ticket_line_has_exactly_one_time
  run a Spec
  each "#21: " line has one stamp before "#21: " and none after it

progress::tests::a_stamped_line_splits_into_its_time_and_message
progress::tests::a_line_without_a_stamp_does_not_split

cargo clippy --all-targets is clean. With cargo test, every test file passes except spec_run: in two full-suite runs, one timing-sensitive concurrency test failed each time, a different one each run. Both passed when run alone, and all 44 spec_run tests passed on a separate run:

• the_parallel_flag_wins_over_spec_parallel: its sessions wait up to 10s for each other, and it failed under full-suite load.
• without_a_limit_at_most_three_tickets_run_at_once: #23: git fetch origin issue-20 failed: fatal: bad object worktrees/widgets-issue-21/HEAD. This is a git race between parallel Tickets sharing the Launch directory, and this change doesn't touch it.

Merge Danger

Door: two-way

Blast Radius: cosmetic

The format of stderr lines changes. A script that matches lines starting with thirdshift: <message> now has to allow for the time. stdout, exit codes and everything posted to GitHub or sent by email stay the same.

Unaddressed findings

Standards

• The HH:MM:SS check is written twice, in src/progress.rs and tests/support/mod.rs: thirdshift is a binary crate, so the integration tests can't call progress::split_stamp. The test copy now has the same name and shape, and its doc comment says which function it mirrors.
• progress::relay both prints and returns the message: removing the prefix and stamp belongs in progress, next to the code that adds them. Splitting it would make spec_run handle the stamp format itself.

Spec

• A message with a newline in it is stamped on its first line only: this already happens with the thirdshift: prefix. Stamping every line would mean changing how multi-line errors print, which the issue doesn't ask for.
• Setup's questions (questions.rs) and the help text after an argument error are printed unstamped: these are interactive prompts and usage text, not progress lines during a Run.
• claude's own stderr passes through unstamped: it is output from claude, not a line thirdshift prints.

Closes #153

🤖 Generated with Claude Code