ci(release): publish any prerelease tag under its own dist-tag

[not-matthias]

Problem

The publish step matched only -alpha in the tag name:

if [[ "${{ github.ref }}" == *"-alpha"* ]]; then
  pnpm publish -r --access=public --no-git-checks --tag=alpha
else
  pnpm publish -r --access=public --no-git-checks   # v5.8.0-beta.1 lands here
fi

Any other prerelease identifier fell through to the else branch and moved the latest dist-tag, so a beta or rc would have been installed by every consumer on ^5. Beta was also unreachable in the first place: scripts/release.sh forwards a single positional to lerna version with no --preid, and lerna defaults the prerelease identifier to alpha, which is precisely why the workflow only ever needed to check for -alpha.

Changes

• Derive the dist-tag from the tag's prerelease identifier rather than hardcoding one channel. beta, rc and next are covered by the same rule, and latest is now reserved for plain vX.Y.Z tags — the fail-closed direction, where an unrecognised suffix gets its own tag instead of clobbering latest. This also drops the loose substring match: v5.8.0-alphabet.1 previously published as alpha.
• Flag the draft GitHub release --prerelease when the version contains a - (this was missing for alpha too).
• scripts/release.sh takes an optional second argument forwarded as --preid, so ./scripts/release.sh prerelease beta produces 5.7.2-beta.0 and the matching tag.

Both shell snippets avoid [[ ... ]] && VAR=... as a final command, since workflow steps run under bash -e where a false test fails the step.

Verification

No release was performed. pnpm, gh and git were replaced with echo-only stubs, and the workflow bodies were extracted verbatim from the YAML file, so these are the argv the shipped text builds:

git tag	publish command
v5.8.0	pnpm publish -r --access=public --no-git-checks
v5.8.0-alpha.0	... --tag=alpha
v5.8.0-beta.1	... --tag=beta
v5.8.0-rc.0	... --tag=rc
v5.8.0-next.3	... --tag=next
v5.8.0-alphabet.1	... --tag=alphabet

• Draft release: 5.8.0 gets no flag, 5.8.0-beta.1 and 5.8.0-alpha.0 get --prerelease. All steps exit 0.
• release.sh argv checked with printf '%q' and an argument count: patch and prerelease both yield 6 arguments with no empty element, prerelease beta yields 8 with --preid beta in position. 0 or 3+ arguments print usage and exit 1.
• bash -n clean, workflow still parses as valid YAML with all four build run-steps intact.

Existing alpha behaviour is unchanged: v*-alpha* still publishes to the alpha dist-tag.

Not included

npm trusted publishing needs no per-channel setup — the OIDC exchange is dist-tag agnostic, and provenance is unaffected. Cross-package deps also already work: workspace:^5.7.1 is rewritten at publish time to ^5.8.0-beta.0, and a caret range whose comparator carries a prerelease matches prereleases of that same version tuple, so @codspeed/vitest-plugin@beta resolves @codspeed/core@beta.

Consuming prerelease versions of the benchmark frameworks themselves (for example vitest@4.0.0-beta.1 against the vitest: "^3.2 || ^4" peer range) is a separate concern and untouched here.

Consistency with the other CodSpeed repos

Beta is the normal prerelease channel elsewhere; codspeed-node was the only repo that could not produce one.

repo	prereleases supported	how it is cut	GitHub release flagged	prereleases historically shipped
runner	yes, first-class	cargo release --execute beta	yes, announcement_is_prerelease from cargo-dist	39 -beta, 24 -alpha tags
pytest-codspeed	yes	uvx bumpver update --minor --tag beta	no	9 -beta tags
codspeed-rust	no	not supported by scripts/release.sh	no	4 -alpha tags, 2023 only
codspeed-node (before)	alpha only, by accident	not reachable through scripts/release.sh	no	2 -alpha tags

Details worth carrying over or noting:

• The runner keeps a prerelease from becoming the default install by gating gh release edit --latest and the downstream CodSpeedHQ/action version bump on announcement_is_prerelease (.github/workflows/post-announce.yml:24-50). npm's dist-tag gives us the equivalent for free, which is what this PR wires up.
• Both the runner (scripts/pre-release.sh:6-10) and pytest-codspeed (scripts/pre-release.sh:6-8) skip changelog generation for alpha/beta/rc. Not applicable here: this repo generates no changelog.
• scripts/release.sh in this repo is a sibling of the codspeed-rust one, which explains the shared <major|minor|patch>-only shape. codspeed-rust has the same gap if it ever needs a prerelease.
• The missing --prerelease on the draft GitHub release is not unique to us: pytest-codspeed and codspeed-rust omit it too.

The CONTRIBUTING.md added here follows the structure of the runner and pytest-codspeed guides, and documents what the prerelease identifier is, since neither of those explains it.

[not-matthias]

@greptileai review

[codspeed-hq]

Merging this PR will regress 5 benchmarks

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 9 improved benchmarks
❌ 5 regressed benchmarks
✅ 224 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Memory	wait 500ms	20.3 KB	64.2 KB	-68.32%
❌	Memory	end	2.1 KB	2.7 KB	-23.43%
❌	WallTime	test_recursive_fibo_10	1.9 µs	2.4 µs	-17.86%
❌	WallTime	fibo darwin	7.8 ms	8.8 ms	-11.72%
❌	Simulation	wait 1sec	57.5 ms	64.3 ms	-10.49%
⚡	Simulation	test_recursive_fibo_10	11,112.5 µs	274.1 µs	×41
⚡	Memory	test sync baz 100	25,080 B	656 B	×38
⚡	Simulation	wait 1sec	57.9 ms	40.7 ms	+42.5%
⚡	Memory	short body	784 B	656 B	+19.51%
⚡	WallTime	switch 2	96 ns	84 ns	+14.29%
⚡	WallTime	short body	2.1 µs	1.9 µs	+11.95%
⚡	WallTime	test sync baz 100	228 ns	204 ns	+11.76%
⚡	WallTime	test sync baz 10	120 ns	108 ns	+11.11%
⚡	Memory	long body	1.3 KB	1.2 KB	+10.39%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing chore/beta-releases (0433753) with main (3735985)}

[greptile-apps]

Greptile Summary

The PR makes prerelease publishing fail closed by deriving and validating an npm dist-tag from the version, while reserving latest for stable releases.

• Marks draft GitHub releases as prereleases when appropriate.
• Documents stable and prerelease release procedures.
• Expands the release helper’s accepted version forms.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
.github/workflows/release.yml	Safely routes prereleases to validated non-latest dist-tags and marks their draft GitHub releases as prereleases.
CONTRIBUTING.md	Documents the lockstep release process and uses the repository-required pnpm tooling.
scripts/release.sh	Documents and accepts explicit prerelease versions while preserving the existing branch and working-tree safeguards.

_{Reviews (4): Last reviewed commit: "fixup! docs: document release bump level..." | Re-trigger Greptile}

[not-matthias]

@greptileai review again