Repository navigation
ci: release ilttu-plugins with release-please - #4
Conversation
release-please keeps one release PR for the ilttu plugin (ilttu-v*) and ilttu-design (ilttu-design-v*). When ilttu-design releases, the Release workflow builds it from the tag, publishes the draft, and opens a PR that pins it in the wrapper with scripts/pin-ilttu-design. The tag-push workflow goes, so one path releases. Check and PR title workflows run the AGENTS.md checks on every PR.
…ugins into ajb/release-please # Conflicts: # AGENTS.md # SECURITY.md
A binary left in ilttu-design/dist made Go stamp every later build of a release vcs.modified=true, as in v0.1.0 and v0.1.1. The build now writes to RUNNER_TEMP and fails if it leaves a file in the checkout. The manifest starts from ilttu-design 0.1.1, now tagged on main, and the plugin's 0.3.1 from the repin PR.
ajbeck
left a comment
There was a problem hiding this comment.
Review of 0251a4b: approve, once this repository is added to the RELEASE_PLEASE_CLIENT_ID and RELEASE_PLEASE_PRIVATE_KEY org secrets' selected repositories. Neither lists it today, so the first Release run would fail when it mints the App token.
Medium (repository setting): main has no branch protection and no rulesets, so SECURITY.md's "pins change only through a reviewed PR" is not enforced; anyone with write access, or the App token, could push to main directly. Add a ruleset that requires a PR on main and lets release-please's tags through.
Low: Check ran at 16:28 and #5 merged at 16:29, so the green run tested a merge ref without #5. The merge is CLEAN, and the v0.1.1 pins survive.
Open for AJ:
- Accept the AGENTS.md rule edits on releases? Recommend yes; they follow ilttu's walle.md.
- Set bootstrap-sha (see the comment on the config)? Recommend yes.
- Turn on immutable releases? Recommend yes, as in ilttu.
- Require PRs on main? Recommend yes.
Verified: every action is pinned by SHA and each SHA matches its tag. Each job's permissions are narrow, and there is no pull_request_target. The secrets are used only to mint App tokens, in the release-please and pin jobs. The build runs from the tag, outside the checkout, and publishes only after the upload. pin-ilttu-design refuses partial input. release-please reads component, draft and force-tag-creation per package. The manifest matches main. No workflow triggers on a tag, so nothing builds a tag twice.
Not verified: the Release workflow has never run; the App's installation here; the first release PR's contents. I made no worktree and no deliberate breaks; the review was read-only.
Start release-please's history at #5's merge, so the plugin's first release PR doesn't repeat what 0.3.1 shipped. Fail the build step when git status fails, download SHA256SUMS in a step of its own so the pin script's step has no token in its env, and let a rerun of the pin job succeed once the pin is on main.
|
Ready for review: #4, head What changed since
|
ajbeck
left a comment
There was a problem hiding this comment.
Review of a2c3f10: fixes accepted, waiting on AJ
All five findings are fixed as suggested:
bootstrap-shaisffd0703.- The comment at :82 no longer tells history.
- The
git statuscheck fails whengit statusitself fails. - A rerun after the pin has merged does nothing.
- SHA256SUMS downloads in its own step, so the pin script runs with no token in its env.
Open for AJ, before merge:
- Add this repository to the
RELEASE_PLEASE_CLIENT_IDandRELEASE_PLEASE_PRIVATE_KEYorg secrets' selected repositories. - The four choices, which the session put to AJ: the AGENTS.md release rules,
bootstrap-sha, immutable releases, and a ruleset requiring PRs on main.
Verified: I read the fix diff from 0251a4b to a2c3f10.
Not verified:
- The Release workflow, which has never run.
- The Check run on this head.
|
AJ's answers to the four choices, in this session: "Yes, all four" (2026-11-15).
Still AJ's before merge: add this repository to the |
#3 is merged, and this now targets
main.Merge #5, the repin to
ilttu-design-v0.1.1, first: the manifest here starts from its versions, ilttu 0.3.1 and ilttu-design 0.1.1.What changed
asteroid-computing/ilttuandorca:release-please-config.jsonnames two packages, and.release-please-manifest.jsonholds their last versions.plugins/ilttu, release typesimple, taggedilttu-v<version>, which bumps$.versionin.claude-plugin/plugin.json.ilttu-design, release typego, taggedilttu-design-v<version>as today, released as a draft with its tag made at once..github/workflows/release.yamlon push tomain:release-pleasekeeps the release PR open, as the org's release App, from theRELEASE_PLEASE_CLIENT_IDandRELEASE_PLEASE_PRIVATE_KEYorg secrets, so its PRs run checks.ilttu-designreleases, a job checks out the tag, vets, tests, builds the four binaries outside the checkout, attaches them withSHA256SUMS, and only then publishes the draft.The old workflow built into
ilttu-design/dist/, so Go stamped every build after the firstvcs.modified=true, in v0.1.0 and v0.1.1 alike; the build now fails if it leaves a file in the checkout.SHA256SUMS, runsscripts/pin-ilttu-design, and opensfix(ilttu): run ilttu-design <version>, which the next release PR ships as a plugin release.scripts/pin-ilttu-design <tag> <SHA256SUMS>rewrites the wrapper'sTAGand four sums, and changes nothing unless every sum is there and every line it rewrites is found..github/workflows/ilttu-design-release.ymlis gone, because a tag the App makes triggers workflows, so it would build and publish the same tag a second time.check.yamlrunsgo vetandgo testinilttu-design/, andclaude plugin validateon the marketplace and the plugin, with Claude Code pinned at 2.1.311;pr-title.yamlchecks a Conventional Commit title of 72 characters or fewer.Why the pins are a PR, not a commit
The wrapper's sums can only be known after the build, and they decide what runs on people's machines, so they change only through a PR someone reviews and merges.
The workflow opens it from the published release's
SHA256SUMS, so nobody copies a sum by hand.Verified
actionlinton all three workflows, and the config and manifest parse as JSON.claude plugin validate .andclaude plugin validate plugins/ilttu, also with an empty config folder and no sign-in, as CI runs them;go vet ./...andgo test ./....scripts/pin-ilttu-designagainst v0.1.0's realSHA256SUMS: it leaves the wrapper as it is, pins a new tag and sum, keeps the file executable, and refuses a shortSHA256SUMS, a tag that isn't ilttu-design's, and a wrapper whose line has changed shape, each without writing.Not verified
main: release-please, the build-and-publish job and the pin job.ilttu-design-v0.1.1is onmain, so release-please should startilttu-designthere; the plugin has no tag, so I expect it to fall back to the manifest's 0.3.1 with a changelog from the start ofplugins/ilttu/'s history.componentfor a package: the schema doesn't list it, thoughsrc/manifest.tsreads it, and without it both packages would tag a barev<version>.Before it can run
RELEASE_PLEASE_CLIENT_IDandRELEASE_PLEASE_PRIVATE_KEYorg secrets' selected repositories; the App is already installed on all of the org's repositories.