/release-kotlin-library

name: release-kotlin-library
description: Use when preparing, publishing, or checking readiness for a Kotlin library release, including verifying its gradle-maven-publish-plugin prerequisite, reconciling changelogs and API snapshots, and checking publication.

Release Kotlin library

Core principle

Publish the prepared, validated, user-approved release commit and call the release complete only after verifying its artifacts and Git state.

Prerequisite

This skill relies on `gradle-maven-publish-plugin` (`com.vanniktech.maven.publish`) for library publication, whether run locally or through tag-triggered CI. API snapshot support assumes Metalava-generated `api/api.txt` files; disable snapshots when the repository does not maintain them.

Procedure

1. Route standalone changelog maintenance to [changelog](../changelog/SKILL.md) and finish that task without applying this skill's publishing prerequisite or release answer forms. For release work, establish scope and inspect repository instructions, Git state, release history, version properties, publishing configuration and required checks. If the user limits review to supplied state, run no commands and mark unsupplied state unknown. Confirm, from build files or supplied repository facts, that published modules apply `com.vanniktech.maven.publish` directly or via a convention plugin; declaration alone is insufficient. If absent or unverified, report it and stop before release mutations; do not install or migrate publishing plugins. Distinguish readiness review, preparation and explicit release authorization; select the matching finish path before proceeding. For recovery or recovery review, select and read the [recovery answer form](references/recovery-output.md) before analysis. For changelog-related review, also load [changelog](../changelog/SKILL.md) in read-only mode. For release preparation, including its changelog portion, read the [preparation answer form](references/preparation-output.md) completely, in its own call, before analysis. Keep reviews read-only, including credentials. Follow the existing local or CI publishing mechanism; do not migrate it. Read the [helper contract](references/helper.md) before configuring the bundled script. Stop before mutation on unsupported layouts or ambiguous destinations. 2. Before preparation, use [changelog](../changelog/SKILL.md) for specified-release notes. If that skill is unavailable, report the missing dependency and stop changelog preparation rather than inventing a substitute. Supply the target version, repository heading conventions and authorized scope; consume its reviewed baseline, notes, coverage or skip result, and gaps. Leave heading finalization to the release helper unless the request is explicitly limited to finalizing the changelog. Resolve coverage gaps before publication. Use supplied release and next-development versions; propose and confirm missing values before release mutation, never silently incrementing a prerelease to a patch snapshot. The caller owns any authorized changelog-correction commit needed before helper preparation's clean-worktree gate. 3. Identify repository release checks, including tests and Metalava API generation and compatibility checks where configured. Confirm that API files are current before snapshotting; the helper copies them without running Metalava. Do not treat other API dump formats as Metalava snapshots. Require passing evidence for the release code; a green parent commit is insufficient after relevant changes. For Gradle execution, use [gradle-run](../gradle-run/SKILL.md), with `--no-scan` unless a scan is explicitly authorized. Fix failed checks within authorized scope; otherwise stop with the failing gate and next action. 4. Configure and preflight the helper with explicit versions, paths, heading style, API snapshot applicability, branch, remote, tag and command arguments. Keep configuration and evidence outside tracked release files. Check the index, snapshot collisions and local/remote destinations before writes. Load `~/.env` for local publication as data, never by shell sourcing or printing it. Preserve explicit process environment values. Missing required credentials block local publication; do not request or log their values. For tag-triggered CI, use existing Git authentication and CI-managed publishing secrets without loading or requiring local dotenv values. 5. Prepare the release: update the version, finalize the changelog heading and applicable published-module API snapshots, run configured checks, and commit only release files. Recheck the finalized changelog against the `changelog` coverage result; return to that skill if finalization changes notes or history. Inspect the resulting commit. Bind validation to this state and invalidate it if relevant code changes. The helper must not publish during preparation. 6. Present the prepared release for explicit user approval before publication: release version and tag, finalized changelog (or its absence), next development version, release commit, artifact coordinates and destination, publishing mechanism, and validation results. Provide the actual notes or a directly reviewable diff, not just a claim that they are ready. Explain that this approval gate is required by this skill and wait for the user's decision; a general request to release does not approve unseen release details. Reuse approval already given for this exact prepared release. If the code, versions, notes or publishing scope change, prepare and validate the revised release and obtain approval again. Preparation-only and readiness requests stop at their requested scope without soliciting publication approval. 7. Once the prepared release is approved and all gates pass, publish from that commit via the