[tonydzi]

v0.3.4 was tagged on 2026-10-31 and never reached PyPI: the release commit changed only CHANGELOG.md, and the publish job stopped on its version check, after the tag. PyPI stayed on 0.3.3 for a week.

This runs the comparison in the normal test suite: pyproject, __init__, the three places in server.json, and the top released CHANGELOG section. A section whose header says unreleased may lead.

Red proof: with pyproject set back to 0.3.3 the test fails and names each place that differs.

🤖 Generated with Claude Code

---

Note

Low Risk
Test-only change with no runtime or release workflow edits; it only adds an earlier guardrail for version bumps.

Overview
Adds tests/test_version_consistent.py so CI fails before merge if release metadata is out of sync—addressing the v0.3.4 case where only CHANGELOG was bumped, the tag went out, and PyPI publish aborted too late.

test_all_versions_agree treats pyproject.toml as canonical and compares __init__.__version__, server.json (top-level version, package version, and the runtimeArguments PyPI pin), and the first CHANGELOG section whose header is not marked unreleased. Mismatches get a single assertion listing each offending location.

Companion tests lock in parsing rules: the broken v0.3.4 changelog-only bump is detected, an unreleased leading section is allowed, headers like v0.3.3.post1 / v0.4.0-rc1 do not count as the stable triple, and a stale == pin in server.json is caught.

^{Reviewed by Cursor Bugbot for commit 1ebdd83. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 5b52867. Configure here.}