PR #1214 release version sync dry-run regression fix
Date: 2026-08-26 (Asia/Shanghai)
Source commit: 3b93a070d607ad9c2dd649b732be14b03041a341

CI first failure (preserved at /tmp/pr1214-version-sync.log):
FAIL: the agent-node baseline really exercises a version difference

Root cause:
The test hard-coded @sleep2agi/agent-node 2.5.0-preview.34. The release bump
made that equal to PAIRED_AGENT_NODE_VERSION, so the dry-run correctly emitted
no WOULD-WRITE line and the test's difference baseline ceased to be a diff.

Fix coverage:
- Derive the next preview independently from each checked-out pair literal.
- Exercise agent-node with its derived differing preview.
- Exercise the checked-out agent-node version as a no-op negative control.
- Retain the stale-alias mutant, witnessed-red old pipefail mutant, and real
  diff-error negative control.

Exact Docker command:
sg docker -c 'docker build --build-arg SOURCE_COMMIT=3b93a070d607ad9c2dd649b732be14b03041a341 -t anet-test1183-pr1214-fix -f tests/test1183-sync-pinned-dryrun/Dockerfile . && docker run --rm --network none anet-test1183-pr1214-fix'

Result:
PASS: normal version differences do not abort the dry-run
PASS: every registered target is followed by the final summary
PASS: same-version negative control proves the version-difference assertion is discriminating
PASS: both release-pair registries own writable quoted literals
PASS: a constant name without a quoted literal is rejected as stale
PASS: WITNESSED_RED: the old diff pipeline exits early and loses the summary
PASS: real diff failures remain nonzero and visible
SOURCE_COMMIT=3b93a070d607ad9c2dd649b732be14b03041a341
RESULT: PASS

Full local output: /tmp/pr1214-test1183-fix.log
