Skip to content

rn-feature-dev

The primary workflow for building features with rn-dev-agent. Takes a feature description and runs an 8-phase pipeline: from codebase exploration through live device verification to PR-ready proof artifacts.

/rn-dev-agent:rn-feature-dev <feature-description>
/rn-dev-agent:rn-feature-dev add a favorites screen with swipe-to-delete
/rn-dev-agent:rn-feature-dev implement dark mode toggle in settings
/rn-dev-agent:rn-feature-dev shopping cart badge that updates in real-time
PhaseNameWhat happens
1DiscoveryParse the feature request, create todo list, ask clarifying questions if needed
2Codebase ExplorationLaunch 2-3 rn-code-explorer agents in parallel to map screens, store, patterns
3Clarifying QuestionsIdentify all ambiguities — error states, navigation, store shape, testIDs — and wait for answers
4Architecture DesignLaunch rn-code-architect agents, produce blueprint with verification parameters and E2E proof flow
5ImplementationBuild the feature following the architect’s build sequence — store, API, components, navigation
5.5Live VerificationProve it works on the simulator — screenshot, health check, component tree, state, interaction, error regression
6Quality ReviewLaunch 2-3 rn-code-reviewer agents for correctness, RN conventions, and project conventions
7SummaryDocument what was built, files changed, decisions made, verification results
8E2E ProofExecute the architect’s proof flow step-by-step, capture screenshots and video, write PROOF.md

This is what separates rn-feature-dev from generic feature development. After implementation, the agent verifies the feature live on the running simulator:

  1. Ensure simulator and navigate to the feature screen
  2. Baseline screenshot and clear error buffer
  3. Health check via cdp_status — Metro, CDP, no RedBox, no errors
  4. Component verification via cdp_component_tree — feature renders correctly
  5. Interaction verification — exercise the primary user action, verify side effects
  6. State verification via cdp_store_state — data shape matches design
  7. Error regression check — no new errors since baseline
  8. Cross-platform comparison — element-by-element iOS vs Android (when both available)

All checks must pass before proceeding to review.

Produces a docs/proof/<feature-slug>/ directory containing:

  • Numbered screenshots for each step of the user flow
  • Video recording with timed step labels
  • PROOF.md with flow table, state snapshots, and deviations
  • PR-BODY.md ready to paste into a pull request
  • Never skip phases — speed comes from parallel agent launches, not phase elimination
  • Ask clarifying questions — identify all ambiguities before implementing
  • Understand before acting — read existing code patterns first
  • Verify on device — prove it works with live screenshots and CDP state checks
  • Cross-platform verification — compare screenshots element-by-element when both platforms are available
  • iOS Simulator or Android Emulator running with the app loaded
  • Metro dev server running
  • For Zustand apps: if (__DEV__) global.__ZUSTAND_STORES__ = { ... } in app entry