Repository navigation
fix(bestax-mcp): release with every bulma-ui release so the published index keeps up - #1027
Conversation
… index keeps up The release job regenerates bestax-mcp's index after bulma-ui releases and before bestax-mcp does (#1018), but bestax-mcp only released on its own commits, so a run that shipped only the library left npm serving an index stamped with an older bestax-bulma. bestax-mcp's release rules now give a patch to every commit that releases bulma-ui, a bulma-ui major included, and the negated scope rule leaves bulma-ui out. The breaking rule sits last because commit-analyzer walks the rules in order and a patch, unlike bulma-ui's major, does not end the walk before a type's `release: false`. scripts/release-rules.test.mjs runs commits through the commit-analyzer semantic-release loads and holds the two packages' rules together. VERSIONING.md states the exception, and CLAUDE.md, CONTRIBUTING.md and bestax-mcp/CLAUDE.md point to it. Fixes #932
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 53 minutes. View limit details
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Preview DeploymentPreview URL: https://559f8fb3.bestax.pages.dev |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Deep review — 0 blocking · 2 advisory
| # | Severity | Area | Finding | Location |
|---|---|---|---|---|
| 1 | 🔵 Advisory | Robustness | PR body's "Listings: nothing applies" contradicts its own "The plugin" section — the A plugin release bullet now fires per bulma-ui release | CLAUDE.md:121 (PR body; no line of its own) |
| 2 | 🔵 Advisory | Correctness | Surviving release prose contradicts the mechanism this PR documents: the non-releasing-types table row, release: false "always wins" in three other configs, and the un-mirrored CONTRIBUTING bullet |
VERSIONING.md:33 |
Overall: The change is sound, and the reasoning in bestax-mcp/release.config.js is correct where it matters most — I read analyze-commit.js and compare-release-types.js at the resolved commit-analyzer@13.0.1 and confirmed that release: false has index -1, so later wins in both directions, which is exactly why the bulma-ui breaking rule has to sit below the release: false types. The riskiest part is not the rules but the blast radius: every bulma-ui release now also cuts bestax-mcp, which cuts a plugin version, which reaches hand-maintained third-party listings — and the PR body tells the maintainer no listing is affected. Focus there first; the rules themselves are the best-tested thing in the diff.
Residual risk:
- Ordering regression in the rule list. Refuted — the new suite bites.
TYPESis the union of commitlint'stype-enumand everytypeeither rule list names, so moving{ breaking: true, scope: 'bulma-ui', release: 'patch' }above{ type: 'build', release: false }flipsbuild(bulma-ui)+ footer frompatchtonullwhile bulma-ui still saysmajor, and the equivalence assertion fails. I verified all five new tests pass (pnpm test:scripts, exit 0, floors held) and that theseenvacuity guard really collectsmajor/minor/patch/null. - A new releasing type added to bulma-ui and not mirrored here. Refuted — covered by the same union plus equivalence assertion; the test does not read the rules, it runs the analyzer semantic-release resolves.
- The negated-glob widening (
!(bestax-mcp)to!(bestax-mcp|bulma-ui)) letting a bulma-ui commit fall through to the angular defaults and over-bump. Refuted by enumeration: the only commitlint type with no rule in either group isrevert, andrevert(bulma-ui)reaches the defaults where neither therevertflag (the angularrevertPatternnever matches the parenthesised form) nor the type matches —nullboth before and after. With a footer it hits the new breaking rule for apatch. - Git-style
Revert "…". Unchanged: it has no scope, so neither glob ever matched it, and{ revert: true, release: 'patch' }in the defaults patch-releases every package exactly ascommitlint.config.jsalready documents. - Unscoped
feat:/fix:would give bestax-mcp a minor, not the "patch whatever bulma-ui's bump" the comment promises — the defaults, not the new group, decide it. Open but not reachable past commitlint'srelease-scope-required, and pre-existing for every package. The new suite covers unscopeddocs:/chore:/chore(release):but no unscoped releasing type. - A bestax-mcp release in a run where bulma-ui does not release. Open, and newly automatic. If bulma-ui's
preparelands the tag andpublishfails, the version is spent (VERSIONING.md); the next push tomainnow releases bestax-mcp alone, andpnpm publishresolves its@allxsmith/bestax-bulma: workspace:^against the bumpedbulma-ui/package.json, pinning a range npm does not serve. Pre-existing shape — anyfix(bestax-mcp)did the same — but this PR guarantees the next push triggers it. The inverse ordering (bulma-ui succeeds, bestax-mcp's step fails) I checked and it is the recovery path:Regenerate the MCP indexreruns and the tarball gets the right stamp. - Downstream frequency. Named rather than refuted:
skills-publish.yml's relay filter (startsWith(github.event.release.tag_name, 'bestax-mcp@')), itsskills-publishgroup withcancel-in-progress: false, and--require-published's bounded backoff all check out as the body claims — but each is now exercised per bulma-ui release, so an npm propagation lag longer than that backoff reds a publish run proportionally more often. Same formcp-registry.ymland Glama's per-release rebuild. - Index freshness on a non-bulma-ui change (a
fix(docs)to an API page, a skill edit, a bestax-migrate class map change) still waits for the next release. Stated in the body and genuinely out of scope here.
🏄 Dude, this one's a clean little drop-in — the rules ride the analyzer's actual walk instead of guessing at it, and the test paddles out and checks every type with and without a breaking footer, so nobody's getting worked by a stale index again. Only thing dragging is the prose on the beach behind it: the listings note says the tide didn't move when the PR's own plugin section says it did. Patch the sign, send it. 🌊
…ch contradicts - VERSIONING.md loses the table row saying the non-releasing types release nothing in any scope. A breaking `build(bulma-ui)` is a bestax-bulma major and a bestax-mcp patch, and the rows and the footer note left behind are all true. - The other release configs no longer say a matching `release: false` always wins. A release type matching after a `false` replaces it, so the claim and the rule it justified both go. Comments only: every config gives the same answers. - The docs-site contributing guide drops the same "don't publish" claim and points at the bestax-mcp patch, and bulma-ui/CLAUDE.md says its releases patch bestax-mcp.
|
deep-review: verify Every open thread has a reply: fixed, with the commit, or refuted with a reason. This settles them on 77047b4. |
Preview DeploymentPreview URL: https://649d76c2.bestax.pages.dev |
There was a problem hiding this comment.
Deep review (verify) — 2 resolved · 0 open
| # | Severity | Area | Finding | Location |
|---|---|---|---|---|
| 1 | 🔵 Advisory | Robustness | PR body's "Listings" section contradicted its own "The plugin" section — verified fixed, the two are now one "The plugin and its listings" section that names the hand-pinned entries | PR body (anchored CLAUDE.md:121) |
| 2 | 🔵 Advisory | Correctness | Release prose contradicting the mechanism (the "(any scope)" table row, "a release: false always wins" in four configs, the docs-site mirror and bulma-ui/CLAUDE.md) — verified fixed by deletion in all three places |
VERSIONING.md:33 |
Overall: This pass settled threads and reviewed no commits. Both of the previous review's advisory threads are addressed in 77047b4b, by deletion rather than by added qualifiers, and both are now resolved. I re-read each site against the current tree and ran scripts/release-rules.test.mjs through pnpm test:scripts: all seven release-rules tests pass, including bestax-mcp patches on exactly the bulma-ui commits that release bulma-ui; the only failures in that run are the documented bulma-ui/dist is absent artefacts of a checkout without a build. While checking thread 2 I also confirmed the one claim the deleted row left exposed — A commit scoped to docs never releases any package — is true, since the suite asserts a breaking feat(create-bestax) gives bestax-mcp null, so a matching release: false does not fall through to the angular defaults.
Residual risk: none identified for this pass — it settles threads only and raises nothing new. A reader who wants the current code reviewed should ask with a deep-review: fresh steer.
🏄 Author caught both advisories clean and paddled back with deletions instead of word salad — the table's tighter, the four configs stopped claiming something the analyzer never did, and the tests still roll in. Zero open threads, good to go.
github-actions
Bot
added
the
review-converged
allxsmith
added
the
needs-human-review
📸 Story screenshots at handoff —
|
github-actions
Bot
added
review-converged
## [1.14.4](https://github.com/allxsmith/bestax/compare/bestax-mcp@1.14.3...bestax-mcp@1.14.4) (2026-10-10) ### Bug Fixes * **bestax-mcp:** release with every bulma-ui release so the published index keeps up ([#1027](#1027)) ([618e483](618e483)) * **bulma-ui:** accept material-symbols 0.47 in the peer range ([539cca2](539cca2)) * **bulma-ui:** drop a Field label's for when nothing it holds takes it ([#1020](#1020)) ([3604c14](3604c14)) * **bulma-ui:** keep a Notification's text clear of its close button ([#1017](#1017)) ([91ebb63](91ebb63)) * **bulma-ui:** keep Theme's component-variable check out of production ([537d90d](537d90d)) * **bulma-ui:** let a caller's aria-label or aria-labelledby name both range Slider thumbs ([047ff7c](047ff7c)), closes [#974](#974) * **bulma-ui:** name a composed DateRangeInputBase's group from a Field label ([75aa6cb](75aa6cb)) * **bulma-ui:** name a range Slider's thumbs from a hand-wired Field label's id ([355065b](355065b)) * **bulma-ui:** name Autocomplete's list from a hand-wired Field label's id ([ad46c06](ad46c06)) * **bulma-ui:** name Autocomplete's suggestion list from a surrounding Field's label ([190183f](190183f)), closes [#998](#998) * **bulma-ui:** name each range Slider thumb from its label ([6b962e3](6b962e3)), closes [#981](#981) * **bulma-ui:** name the date and time picker bases from a Field label ([e7ae797](e7ae797)) * **bulma-ui:** say a label placed in Field.Label reaches no thumbs or list ([07c2a99](07c2a99)) * **bulma-ui:** say FieldLabel renders the label column, not a label ([2a6ab10](2a6ab10)) * **bulma-ui:** say the label prop needs wiring to reach thumbs or list in an inner Field ([9b33d8d](9b33d8d)) * **bulma-ui:** say which controls a hand-wired Field label's id reaches ([514127f](514127f)) * **bulma-ui:** warn when Theme is given a variable Bulma sets on the component ([bb8f9b8](bb8f9b8)), closes [#1021](#1021) * **create-bestax:** announce the starter's notifications through a status region ([91dbfd5](91dbfd5)) * **create-bestax:** give each starter notification a status region of its own ([f1ef4df](f1ef4df)) * **create-bestax:** keep focus on the counter when Reset disables itself ([733e505](733e505)) * **create-bestax:** keep the reason for the stylesheet order in every scaffolded entry file ([f4591e0](f4591e0)) * **create-bestax:** pin material-symbols to the newest range bestax-bulma accepts ([ff4f820](ff4f820)) * **create-bestax:** return focus when the starter's notification closes, and tighten guards ([67e3678](67e3678)) * **create-bestax:** say which helper props render nothing under the no-helpers flavors ([b29985b](b29985b)) * **create-bestax:** stop the no-helpers CLAUDE.md offering helper props as the way out ([a96c203](a96c203)) * **create-bestax:** stop the vite-ts build shadowing its config, and fix the starter page ([1dd964d](1dd964d)) * **docs:** give the mixin note the modifier caveat its lead carries ([1427f08](1427f08)) * **docs:** keep the modifier caveat for mixin variables in the MCP index ([e227663](e227663)) * **docs:** name the one edge of the rule for which variables Theme reaches ([b33ebfa](b33ebfa)) * **docs:** qualify the guide and the theming skill on what Theme reaches ([9c2347b](9c2347b)) * **docs:** read only a component's own mixin as its home ([f40cd2c](f40cd2c)) * **docs:** refuse a CSS-variable scope the page cannot fully word ([c815602](c815602)) * **docs:** say where component-scoped Bulma variables have to be set ([771d5bd](771d5bd)), closes [#1021](#1021)
|
🎉 This PR is included in version 1.14.4 🎉 The release is available on: Your semantic-release bot 📦🚀 |
# [5.28.0](https://github.com/allxsmith/bestax/compare/@allxsmith/bestax-bulma@5.27.8...@allxsmith/bestax-bulma@5.28.0) (2026-10-12) ### Bug Fixes * **bestax-mcp:** release with every bulma-ui release so the published index keeps up ([#1027](#1027)) ([618e483](618e483)) ### Features * **bulma-ui:** add HoverCard, a card that opens on hover or keyboard focus ([#1030](#1030)) ([0eb41df](0eb41df))
|
🎉 This PR is included in version 5.28.0 🎉 The release is available on: Your semantic-release bot 📦🚀 |
1 participant
#1018 fixed the first half of #932: the publish job now regenerates bestax-mcp's index after bulma-ui releases and before bestax-mcp does, so a run that releases both ships an index stamped with the library it just released. That only helps when bestax-mcp actually releases, though, and it only ever released on its own commits. Any run that shipped just the library left npm serving the old index. Right now the newest bestax-mcp on npm, 1.14.3, ships an index stamped bestax-bulma 5.27.2, and bestax-bulma is on 5.27.8.
So this makes bestax-mcp cut a patch every time bulma-ui releases. The index documents the library, so it should ship when the library does.
Why it's config and not a workflow change
The job already does the hard part in the right order. What was missing was a decision about when bestax-mcp releases, and that decision already has a home in
bestax-mcp/release.config.js. Doing it in the workflow would have meant a second place that decides releases, plus a step that either fakes a commit or works around semantic-release, sitting in the job that holds the bypass token. Nothing under.github/changes here.How the rules work
{ scope: '!(bestax-mcp|bulma-ui)', release: false }, so it never meets a bulma-ui commit.feat,fix,perf,refactorandstyle, which are the types bulma-ui releases on, plus any breaking bulma-ui commit. It's a patch whatever bulma-ui's own bump, a major included. The server's tools and prompts don't change, only what they answer from, and a major would strand everybestax-mcp@1install and listing.The part that surprised me: the old comment said a matching
release: falsealways wins. It doesn't quite. commit-analyzer walks the matching rules in order, afalsereplaces whatever came before it, a release type after afalsereplaces thefalse, andmajorends the walk. That's whychore(bulma-ui)with aBREAKING CHANGE:footer releases a bulma-ui major today even though{ type: 'chore', release: false }matches it: the major ends the walk first. Here the breaking rule gives a patch, which ends nothing, so it has to sit below the non-releasing types or a breakingbuild(bulma-ui)would release a library major and no index. The config says all of this next to the rules.Two more things fell out of that. The glob change isn't what makes bulma-ui commits release here, since the bulma-ui rules come later and would beat that
falseanyway. What it buys is that the rule's place in the list stops mattering. And the glob change without the new rules would be worse than doing nothing: bulma-ui commits would fall through to the angular defaults and give bestax-mcp a minor or a major.How I checked it
scripts/release-rules.test.mjsruns commits through the actual commit-analyzer, the copy semantic-release resolves for itself, rather than reading the rules. It takes every type commitlint accepts, with and without aBREAKING CHANGE:footer, scoped to bulma-ui, and holds bestax-mcp to a patch exactly when bulma-ui releases. It also checks mixes, bestax-mcp's own bumps, other scopes, the release job's ownchorecommits, and that the link only runs one way.To make sure the test bites, I mutated the config and ran the test against each change: the breaking rule moved above the non-releasing types, a bulma-ui rule dropped, one set to
minor, the whole group dropped, and the old glob moved to the end of the list. Every one fails. The one mutation that passes is putting the old glob back in its old spot, which gives the same answers.The same commits through both versions of the config, with bulma-ui's unchanged config for comparison:
feat(bulma-ui)fix(bulma-ui)feat(bulma-ui)+BREAKING CHANGE:build(bulma-ui)+BREAKING CHANGE:docs(bulma-ui),chore(bulma-ui)fix(create-bestax),docs:,chore(release)feat(bestax-mcp)feat(bestax-mcp)+BREAKING CHANGE:feat(bulma-ui)+BREAKING CHANGE:andfix(bestax-mcp)feat(bulma-ui)andfix(create-bestax)anddocs:I also ran the real commits on main since
bestax-mcp@1.14.3. With main's config there's no release pending, with this branch's there's a patch from the bulma-ui fixes that already shipped, and bulma-ui's answer doesn't move. This commit isfix(bestax-mcp)too, so merging it releases bestax-mcp right away with the current index.The changelog
release-notes-generator already lists
feat,fixandperfcommits from every scope. bestax-mcp's CHANGELOG has bulma-ui and bestax-migrate entries in it today, so I left release-notes-generator as it is. What's new is that some bestax-mcp releases will be nothing but bulma-ui commits, and a breaking bulma-ui commit lands under BREAKING CHANGES in a bestax-mcp patch. I generated notes for a sample range to see it. It reads a little odd, but those are the changes the new index describes, and the config comment says so.Prose the change made stale
The other release configs said a matching
release: falsealways wins, which is the claim that makes the ordering above look unnecessary. They now say only that every rule is checked, comments only, and every config gives the same answers. VERSIONING.md also loses the table row that saiddocs,chore,ci,testandbuildrelease nothing in any scope, since a breakingbuild(bulma-ui)is a library major and now a bestax-mcp patch too. The docs-site contributing guide drops the same claim and points at the bestax-mcp patch, andbulma-ui/CLAUDE.mdsays its releases patch bestax-mcp.The plugin and its listings
The
bestaxplugin pins bestax-mcp, andskills-publish.ymlrelays everybestax-mcp@release into a publish run on main. Its version counts commits that touchbestax-mcp/package.json, so each bestax-mcp release also bumps the plugin. More bestax-mcp releases means more plugin publishes and more update prompts for plugin users. Nothing in that flow breaks: the relay already filters tobestax-mcp@tags, publish runs share a concurrency group that never cancels one midway (a newer pending run replacing an older one loses nothing, since it checks out main later), and--require-publishedretries a 404 for about a minute before refusing a pin npm doesn't serve yet. The restamp commit only touchesdata/catalog.json, which isn't plugin content, so it adds no bump of its own. The MCP Registry gets a listing per release the same way it does now, and Glama already rebuilds on every GitHub release in the repo.So the distribution guide's "A plugin release" bullet now applies to every bulma-ui release. The entries it names pin the plugin version by hand (awesome-copilot, the OpenAI directory, Kiro, Cursor Marketplace and Build with Claude), so each one that lists the plugin falls a release behind every time bulma-ui ships, and nothing in CI can see it. The bullet already says to update them when a release should reach them, and that stays a call per release rather than a chore for every bulma-ui PR. None of the other "What goes stale" triggers apply: these are patches, no skill or tool changes, and the start command and major stay put.
What this leaves alone
Index changes that don't come from a bulma-ui release, like a
fix(docs)to an API page's usage section, a skill edit, or a bestax-migrate class map change, still wait for the next bulma-ui or bestax-mcp release. In practice that's usually the next bulma-ui release now.Checked with the new test,
test:scripts(coverage floors hold),check:conformance, prettier and eslint.Fixes #932