[cdalar]

Fixes the third cmd/mux.go bug from the review that went with #1275 (#1 and #2 were fixed in #1276).

The bug

Two onctl claude runs starting the same herdr session at once both found no workspace, and each made one. That left:

• Two Claudes in one directory, under one label.
• A session that couldn't be attached. ws held two ids, so herdr workspace focus "$ws" failed on every attach. running() still reported true, so retrying never got past it, and the only fix was closing a workspace by hand.

A related race (review finding #4): one run's server restart could close the workspace another run had just made.

Fix

• Lock: herdr's start script, from the server check to pane run, now runs under flock on /run/boxctl/herdr.lock, which serializes concurrent starts on a box. The lock fd is closed before exec herdr, so attaching doesn't hold it. The background server doesn't inherit it (9>&-).
• No flock: where flock is missing, the script runs without the lock, as before. It is part of util-linux on Debian and Ubuntu, and busybox has it.
• Existing duplicates: a label already on two workspaces now attaches to the first instead of failing.

Tests

• TestHerdrStartConcurrent: two starts race against a slowed-down fake herdr, and exactly one workspace is made.
  • It skips on macOS, which has no flock.
  • I ran it on Linux, cross-compiled and run in a golang:1.26 container: it passed 3 out of 3 runs, and with the lock removed it fails with "2 workspaces made for one session". CI runs on Ubuntu, so it runs there.
• TestHerdrStartDuplicateLabel: with a label already on two workspaces, the run focuses w1 and makes no new workspace.
  • The fake herdr's workspace focus now rejects unknown ids, as real herdr does. Without that, the test passed even with the fix removed.
• Checks: go test ./cmd/ passes and golangci-lint reports 0 issues.

Not tested against a real herdr on a box.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Ba6Vc66D5hH1xSTvJcCbVz

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 50.68%. Comparing base (5b34c3f) to head (2006002).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1278      +/-   ##
==========================================
+ Coverage   50.67%   50.68%   +0.01%     
==========================================
  Files          83       83              
  Lines        9840     9843       +3     
==========================================
+ Hits         4986     4989       +3     
  Misses       4854     4854              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.