All seventeen workflows are first-class on every host:
Claude / Cursor: /rn-dev-agent:<workflow> [arguments]
Codex: $rn-dev-agent:<workflow> [request text]
Codex also has eleven domain skills, for exactly 28 native skills. Its workflow
skills are explicit-only and best-effort source-command-* migration is
disabled. The primary workflow is rn-feature-dev; the same names below apply
on every host.
Command When to use run-workflowEstablish the proven operating sequence before any real device journey rn-feature-devBuild a complete feature with the 8-phase pipeline test-featureTest an already-implemented feature end-to-end qa-prDevice-test a GitHub PR on simulator, emulator, and/or physical device build-and-testBuild the app first, then test debug-screenDiagnose a broken screen — gather evidence, apply fix observeWatch the agent live in a web UI — timeline, device mirror, state panels, browser-run actions
Command When to use list-learned-actionsList saved actions, flows, and feedback memories run-actionReplay a saved action with self-repair lock-e2ePromote a verified action into a frozen, locked e2e regression test
Command When to use proof-captureRecord video + screenshots for PR proof nav-graphMap and navigate the app’s screen graph
Command When to use setupOnboard a project — Claude manages CLAUDE.md; Codex manages AGENTS.md; writes are previewed doctorStrictly read-only host/package/MCP/schema/task/environment diagnosis check-envVerify Metro, CDP, and simulator are ready check-vercel-rulesAudit the project against the bundled best-practice rules
Command When to use send-feedbackReport a bug or send feedback via GitHub issue