Watch and Resume Benches¶

This page is for debugging

In practice the orchestrator watches the benches for you. The follow loop reads their state as pull requests are published and merged, and the orchestrator tells you which one needs you: a pull request to review, a failed bench, a stuck set. The commands here show how a bench works, and what to reach for when one misbehaves and you want to look yourself; you do not need them to run a project.

Watch a running bench¶

Benches run detached, one process per task, with no terminal attached. Each writes everything it does to .logs/<task-id>.log in the worktree root, beside the worktrees and outside every repository, so a log survives the removal of its worktree (--log-dir moves it).

A bench's log prints REFINE when its second session starts, PUBLISHED with the pull request URL when it ends well, and REFINE-FAILED if the review session failed and the change was published as the first session left it.

benches=${ETABLI_WORKTREE_ROOT:-../benches/$(basename "$PWD")}
tail -f "$benches/.logs/a-001.log"       # follow one bench
tail -n 40 "$benches"/.logs/*.log        # where each bench got to
pgrep -af 'tasks/.*\.md'               # which benches are still alive

The worktree is the other live view. The commands below name it once, from the planning repository, wherever the launcher put it:

home=$PWD
bench=${ETABLI_WORKTREE_ROOT:-../benches/$(basename "$home")}/a-001
git -C "$bench" status                  # what the bench has changed
git -C "$bench" log --oneline           # what it has committed

That is usually a faster answer than reading its log.

Stop a bench¶

To stop a bench, kill its process, then remove the worktree and the claim:

pkill -f 'tasks/a-001\.md'          # a bench is known by its task file
git worktree remove --force "$bench"
git push origin --delete task/a-001

Deleting the remote branch is what releases the claim. Skip it and the task stays held by you and nobody else can pick it up.

A bench whose task names a repo belongs to that clone, not to the planning repository: its worktree is registered there, and its claim is on that repository's origin. Run the last two commands against the clone, for repo: shop-iac:

git -C ../shop-iac worktree remove --force "$bench"
git -C ../shop-iac push origin --delete task/a-001

Pick a bench back up by hand¶

You cannot type into a running bench: -p sessions have no input attached. What you can do is resume the conversation once it has finished or stopped, in the same worktree, with its context intact.

Copilot benches print the command to do it as their last line, so it is already in the log:

grep -h 'copilot --resume' "$benches/.logs/a-001.log"
cd "$bench" && copilot --resume=<session-id>

Claude benches do not print a session id in text mode, so there is nothing in the log to resume from. Use claude --continue from inside the bench worktree, which picks up the most recent session for that directory:

cd "$bench" && claude --continue

That is reliable only while the bench is the last session started in that worktree. When it is not, start a fresh session there and point it at the task file and the branch:

cd "$bench" && claude "/etabli-implement $home/specs/tasks/a-001.md"

To redo only the review of a change the first session already left in the worktree, run /etabli-refine on the same task file and --report .bench-report.md instead.

Claude Code can also run benches as attachable background sessions (claude --bg, then claude attach <id>), which would remove this asymmetry on the Claude side. The launcher does not use it: Copilot CLI has no equivalent, the logs come back as raw terminal output, and a background session waits for input instead of exiting, so a stuck bench would hold its claim indefinitely.