xilantra/skills

Reusable Agent Skills by Afiq Xilantra Azmi.

Included skill

cross-platform-app-port

Ports an existing app, feature, or user flow from one platform or framework to another while preserving verified product behavior and producing a maintainable, target-native implementation.

The skill supports workflows such as:

• iOS to Android
• Android to iOS
• Web to native mobile
• Electron to native desktop
• UIKit to SwiftUI
• Views to Jetpack Compose
• React Native or Flutter to native platforms

Bird's-eye view

flowchart LR
    A[Existing product] --> B[Define port scope]
    B --> C[Characterize verified behavior]
    C --> D[Separate product truth from source debt]
    D --> E[Map target-platform equivalents]
    E --> F[Build one target-native vertical slice]
    F --> G[Build, test, capture, compare]
    G --> H{All gates pass?}
    H -- No --> I[Classify the gap]
    I --> F
    H -- Yes --> J[Record evidence and migration state]
    J --> K{More in-scope slices?}
    K -- Yes --> F
    K -- No --> L[Human approval before merge or release]

Loading

The loop may run interactively or within explicit limits for long-running work. It must stop for blockers, exhausted budgets, iteration limits, or actions that require human approval.

Install this skill

npx skills add https://github.com/Xilantra/skills --skill cross-platform-app-port

Install all Xilantra skills

npx skills add xilantra/skills

The Skills CLI installation pattern is documented by skills.sh.

Validate

Run the bundled structural validator:

python cross-platform-app-port/scripts/validate_skill.py cross-platform-app-port

Validate the migration-state template:

python cross-platform-app-port/scripts/validate_migration_state.py \
  cross-platform-app-port/assets/MIGRATION_STATE.template.json

Then run the official Agent Skills reference validator when available:

skills-ref validate ./cross-platform-app-port

Development status

Version 0.2.0 adds workflow scaling, bounded and multi-agent loop controls, commit-bound verification evidence, human approval gates, and bird's-eye workflow diagrams. The included eval cases still need real-agent runs with and without the skill before claiming a measured pass rate.

License

MIT. See LICENSE.