[rasros]

What changed

• Keep CI in two identical workflows, build.yml and release.yml, across Kpermute, Skema and Kencode. Replace Python version routing and simulator selection with inline Bash and jq.
• Remove verifier scripts, their tests, the setup composite action, staging checks and independent consumer fixtures. Routine CI runs one Gradle check/lint/coverage graph.
• Compile and test the libraries on JDK 17 only. Gradle runs on JDK 21 because the published kbuild 1.3.6 plugin requires it; remove both additional packaged-JAR test runs and JDK 25 setup.
• Preserve default JVM/Linux x64 targets, full target opt-in, parallel Linux/web/Windows/macOS/iOS release validation, encrypted configuration caches, main snapshots and tag-only releases. Only stable tags publish to Central; RC tags publish to GitHub only.

Why

The small libraries need straightforward Gradle builds without a separate publication-verification framework.

Testing

• Workflow lint and ShellCheck pass; 15 temporary Bash version-routing cases pass, including numeric tag ordering, malformed tags, exact RC versions and rejection of snapshot tags.
• Temporary checks confirm simulator selection, JDK setup, RC publication conditions and byte-identical workflows across all three libraries.
• Kpermute's JDK 17 check/lint/coverage graph passed locally using the required JDK 21 Gradle runtime. Kpermute also assembled all 12 RC publications locally, and all 65 asset checksums passed without publishing.
• Routine CI passed. Full platform CI passed Linux/web, Windows, macOS ARM64 and iOS simulator checks.
• Successful full Linux logs confirm the encrypted configuration cache was saved.