[gering]

Two fixes in plugins/work-system/scripts/mandate.sh.

1. A rebase over main's neighbouring lines keeps coverage

On #89, a rebase flipped covered to no (changed-since-review) although the lane's own lines were unchanged. lane_diff_id hashed a -U3 diff, whose three context lines per hunk come from main.

• The id now hashes a -U0 diff (measured: -U0 alone is enough; stripping context after -U3 would keep -U3's hunk merging). A hunk's place is its file, its order among the file's hunks, and git's function-context line. Line numbers are not part of it, because any main insertion above would shift them.
• Still flips to no: a lane line that changes, or that moves under another header or past another lane hunk. A main edit to the hunk's function-context line still fails closed.
• Accepted residuals (raised in the local review, see below): a lane line moved within one function past no other lane hunk keeps the id, and so does a main edit that changes what a lane hunk means (a guard deleted right above it).
• Compat: reviews recorded by 1.28.0 (diff_id_form=context, masked -U3) or earlier (legacy) still match their unchanged lane, with the old argv, so this also holds under a non-default diff config. The tests recompute both old ids from the old algorithms. delta still shows context; diff_form=masked is now context-free.

2. acquire_round_lock TOCTOU (CI flake, rc=4)

After a failed mkdir, the lock code tested [ ! -d "$lock" ] to tell "cannot create" from contention. The holder can release the lock in between, so a concurrent booking exited 4. Reproduced under parallel load: 16 of 72 runs failed.

• The verdict now comes from mkdir's own error (File exists = contention). The fix passed 168 runs under load with 0 failures.
• The pid is written aside and renamed in, so lock_is_stale never reads a partial pid. A young lock without a pid is never stale, which covers the mkdir-to-pid gap.
• A file or symlink at the lock path now exits 4 at once with a hint.
• Deterministic test: a PATH mkdir shim that fails once with "File exists" and leaves no lock behind.

Review

Local /swarm:review (Claude + gpt-6.1-sol + grok-4.7; 2 grok clusters timed out): 11 findings. Agreed fixes are applied and recorded (covered=yes via=agreed-fixes). Not changed:

• A 1.28.0 session cannot read a 1.28.1 record. This is fail-closed version skew, which /sessions fixes.
• The test oracle's version regex only needs to match the tested versions.

Checks

• test_mandate.py: all tests pass
• scripts/check-structure.py: OK (pre-existing pr-flow /open description warning)

work-system 1.28.0 → 1.28.1 (patch).

🤖 Generated with Claude Code

https://claude.ai/code/session_01BJZ4tA8d432yqatzZMJqFw