skainet-coding-skills

Official Claude Code plugin marketplace for working with SKaiNET — the Kotlin Multiplatform on-device ML/AI framework. Maintained under the SKaiNET-developers organization.

The marketplace hosts two side-by-side skill bundles. They're designed to be installed together; per-skill descriptions disambiguate via path heuristics so exactly one fires per prompt.

Plugins

skainet-contributor-skills

For developers contributing to SKaiNET itself (editing files inside the SKaiNET repo).

Skill	What it covers
kotlin	Explicit-API mode, package layout, sealed hierarchies, value class.
gradle-multimodule	Version-catalog discipline, convention plugins, BOM membership, binary-compatibility-validator.
kmp	Full target matrix (JVM/Android/iOS/macOS/Linux/androidNative/JS/WASM), source-set hierarchy, KSP wiring.
skainet-java-interop	The @file:JvmName / object + @JvmStatic facade convention under */jvmMain/kotlin/sk/ainet/java/.
skainet-testing	Kotest in jvmTest, kotlin.test in commonTest, TensorAssertions + ToleranceConfig, JUnit 5 mirror in skainet-test-java.

skainet-consumer-skills

For developers using SKaiNET as a library in their own Kotlin / Java / Android / KMP app.

Skill	What it covers
skainet-consumer-setup	BOM (sk.ainet:skainet-bom), library coordinates (sk.ainet.core:skainet-*), artifact picker.
skainet-data-dsl	tensor { shape(…) { … } }, transform pipelines, slice views.
skainet-nn-dsl	sequential<T, V> { } and dag { } neural-network builders.
skainet-model-loading	GGUF, ONNX, SafeTensors, JSON loaders.
skainet-inference	ExecutionContext, model.forward(x, ctx), batching, threading, TurboQuant KV-cache.
skainet-android-integration	assets/ → cacheDir, viewModelScope + Dispatchers.Default, ABI filters, onTrimMemory.
skainet-java-consumer	Maven dependencies, SKaiNET.context(), TensorJavaOps.*, StableHloConverterFactory, TokenizerFactory.

Installation

In any Claude Code session:

/plugin marketplace add SKaiNET-developers/skainet-coding-skills
/plugin install skainet-contributor-skills
/plugin install skainet-consumer-skills

Or clone locally and point at the working tree:

git clone https://github.com/SKaiNET-developers/skainet-coding-skills.git
# then in Claude Code:
/plugin marketplace add /path/to/skainet-coding-skills

Repository layout

.
├── .claude-plugin/
│   └── marketplace.json                 # lists both plugins
├── skainet-contributor-skills/
│   ├── .claude-plugin/plugin.json
│   └── skills/
│       ├── INDEX.md
│       ├── kotlin/
│       ├── gradle-multimodule/
│       ├── kmp/
│       ├── skainet-java-interop/
│       └── skainet-testing/
└── skainet-consumer-skills/
    ├── .claude-plugin/plugin.json
    └── skills/
        ├── INDEX.md
        ├── skainet-consumer-setup/
        ├── skainet-data-dsl/
        ├── skainet-nn-dsl/
        ├── skainet-model-loading/
        ├── skainet-inference/
        ├── skainet-android-integration/
        └── skainet-java-consumer/

Each skill directory contains a SKILL.md (frontmatter + workflow + canonical examples), a references/ folder with lookup tables, and an evals/evals.json with first-pass scenario tests.

Conventions

• Every code example in a SKILL.md ends with // from: <path>:<line-range> so the citation check (CI on every PR) verifies it stays in sync with SKaiNET source as the framework evolves.
• Descriptions explicitly say when each skill should NOT fire, to prevent trigger collisions across the two plugins.
• Constraint skills carry numbered hard rules; the two pure-DSL skills (skainet-data-dsl, skainet-nn-dsl) lead with a cheatsheet instead.

Contributing

See CONTRIBUTING.md for skill authoring, citation rules, and the eval scenario format.

Related

• SKaiNET — the framework this marketplace serves.
• SKaiNET website
• DeepWiki for SKaiNET
• Claude Code plugins — how the marketplace and skills format work.

License

MIT — see LICENSE. Matches the upstream SKaiNET license.