Re-land #341 (openBenches handoff environment) in v1.2.1, with its review's fixes #343

Re-land #341 (OPENBENCHES_CLAUDE_* with the WORKBENCHES_CLAUDE_* fallback in both handoff surfaces) in v1.2.0, with the fixes its review owes.

How it got here:

• feat: accept openBenches handoff environment #341 was squash-merged into main as ee07c0e at 2026-10-25T20:52:07Z.
• Brett then ruled to reset main to 33ce26d, so that the v1.1.0 cutover keeps its approved source (Triad migration (feature 004-migrate-to-triad): execution tracking under Brett Heap's word of 2026-10-07 #186 6414151196). Lane 3 reset it on 2026-10-26 at 11:05Z (Triad migration (feature 004-migrate-to-triad): execution tracking under Brett Heap's word of 2026-10-07 #186 6414491718).
• So feat: accept openBenches handoff environment #341's change is not on main. Its pull request still shows as merged, and its commits stay reachable through it.

The branch: 033-openbenches-compatibility at 4e687e2f, with two commits on 33ce26d:

• 576b39f, "feat: accept openBenches handoff environment";
• 4e687e2, "fix: avoid eval in Claude environment checks".

What the re-land must add before it merges:

• From lane 5's independent review (feat: accept openBenches handoff environment #341 6401046464, READY AFTER FIXES):
  • S2: every test scrub that clears WORKBENCHES_CLAUDE_* also clears OPENBENCHES_CLAUDE_*, at the places the review lists. Without that, a host's OPENBENCHES_CLAUDE_LANE_DIR leaks into test_restart_refuses_unsupported_checkout_or_agent_before_preservation[none].
  • S3: tests for:
    • a name set to an empty value;
    • the lifecycle state root in the conflict test's snapshot;
    • where the block sits in the skill;
    • a conflict under --restart, --exit, --in-process and --late.
• From lane 3's ruling on S1 (feat: accept openBenches handoff environment #341 6401064898):
  • a conflict refuses only where the variable is read, before any write;
  • a conflict in a variable the run does not read prints one line and does not refuse.
• N1 (Sourcery's eval finding) is answered by 4e687e2.

When: after v1.1.0 is tagged, in the landing order after #303, which changes the same two files. An independent re-check of the fixes, and a merge-tree against #303's branch, come before it lands.

Lane: openRepoTools-3