netlift-sdk

NetLift setup skill

netlift-native-app helps a coding agent create or connect native Kotlin/Android and Swift/iOS apps to shared .NET code. It supports new or existing projects on either side, reviews the reachable public API, suggests compatible facades, and verifies actual runtime results.

Install

Copy the netlift-native-app folder into your agent’s supported skills directory. For a repository-local Copilot skill, run from the consuming repository:

mkdir -p .github/skills/netlift-native-app
curl --fail --location \
  https://raw.githubusercontent.com/Redth/netlift-sdk/main/skills/netlift-native-app/SKILL.md \
  -o .github/skills/netlift-native-app/SKILL.md

Inspect the downloaded file before using it. If a skill with that name already exists, compare it rather than overwriting local customizations. Other agents can use the same standard SKILL.md in their supported skill location; no EngrAIm installation or bundled executable is required.

Example requests

The skill is a setup workflow, not a claim that every mobile platform is supported. It distinguishes in-memory Swift previews from C# execution and must surface unsupported targets or missing hosting steps.

Validation and maintenance

The skill passed its structural/safety gate before export. Its package pins and commands were checked against public-feed Android alpha.14 and Swift alpha.9 consumers. The README’s new-project C#, Kotlin, and Swift examples were extracted into a separate project and built; the Android app returned the real managed greeting and the Swift preview returned its explicitly in-memory greeting.

The alpha.14 rerun used empty Gradle and NuGet caches, then verified the managed greeting and all ten signed R8 Orders acceptance markers twice across process relaunch. Swift alpha.9 generation also matched the real source-owned iOS workload host’s contracts and passed its managed acceptance separately.

The recipe also covers the checked-in existing-project Orders consumers: runtime-default clients, explicit injection, installation errors, callbacks, streams, and cancellation. It is not an automated multi-agent benchmark; recheck the canonical public docs and package metadata when updating it.