Create agent skills

[thymikee]

Based on the documentation for the tools in this repository, create agent skill(s) to help agents use the CLI and RN library

[californiandreamer]

Hi! I pushed a proposed skill architecture in PR #127.
This PR covers the react-native-legal implementation flow for the bare React Native workflow.
I’d like to confirm this architecture early before scaling to Expo/Node and router/troubleshooting skills.

In parallel, I’m benchmarking how much the agent improves bare-workflow implementation quality.

Goal

Handle full react-native-legal implementation to agents by as simple as possible prompt based on agents skills.

Current bare implementation skill

Flow

• Trigger when prompt asks to integrate/display OSS licenses in a bare React Native app.
• Install dependency in app workspace react-native-legal.
• Generate native legal metadata from app root: yarn react-native legal-generate.
• Choose exactly one UI mode:
  • native list (ReactNativeLegal.launchLicenseListScreen(...)), or
  • custom list (ReactNativeLegal.getLibrariesAsync()).
• If preference is ambiguous, ask one focused question.
• If prompt asks for a new settings screen, default to custom list mode.
• Avoid mixing native + custom in one implementation unless explicitly requested.
• Validate artifacts:
  • Android android/config
  • iOS Settings.bundle/Root.plist
  • iOS Generate licenses with LicensePlist build phase
• Use deterministic check: node skills/integrating-rn-legal-bare/scripts/verify-bare-artifacts.mjs --root <app-root>.

Skill architecture

Current bare skill package (canonical source):

skills/
└── integrating-rn-legal-bare/
    ├── SKILL.md
    ├── agents/
    │   └── openai.yaml
    ├── references/
    │   ├── commands.md
    │   ├── decision-tree.md
    │   ├── implementation.md
    │   ├── known-limitations.md
    │   └── validation-checklist.md
    └── scripts/
        └── verify-bare-artifacts.mjs

Runtime exposure:

• .agents/skills/integrating-rn-legal-bare/** -> symlinks to skills/integrating-rn-legal-bare/**
• .claude/skills/integrating-rn-legal-bare/** -> symlinks to skills/integrating-rn-legal-bare/**

Final implementation proposal

After validating the bare workflow skill, the model scales to a router-driven multi-workflow architecture:

flowchart LR
  A["User Prompt + Repo Context"] --> B["Router: routing-rn-legal-workflows"]

  B --> C["Detect Intent Signals"]
  B --> D["Detect Repo Signals"]
  C --> E{"Route Decision"}
  D --> E

  E -->|Expo| F["Skill: integrating-rn-legal-expo"]
  E -->|Bare RN| G["Skill: integrating-rn-legal-bare"]
  E -->|Node / CLI / API| H["Skill: analyzing-licenses-node"]
  E -->|Conflict/Unclear| I["Ask one focused clarification"]

  F --> J["Expo Implementation + Validation"]
  G --> K["Bare Implementation + Validation"]
  H --> L["Node License Analysis + Validation"]

  J --> M{"Any errors/constraints?"}
  K --> M
  L --> M

  M -->|Yes| N["Skill: troubleshooting-rn-legal"]
  M -->|No| O["Final Output"]

  N --> P["Targeted fix + re-validation"]
  P --> O["Final Output"]

  O --> Q["Benchmark Scoring via evals/skills"]

Loading

Tasks

• Test baseline behavior with evals (Codex + Claude)
• Prepare single workflow skill + router base, push draft
• Add rest workflows skills:
  • Expo
  • Node
• Add helper skills
  • Routing
  • Troubleshooting
• Test final skill behavior with evals (Codex + Claude)