test-feature

Test this React Native feature: $ARGUMENTS

Run the rn-tester protocol INLINE (parent session)

Important (GH #31): Do NOT spawn the rn-tester agent via the Task tool. MCP tools (cdp_*, device_*) are not available in spawned subagents. Execute the rn-tester protocol directly in this parent session using the rn-testing and rn-device-control skills as your reference.

Load the rn-testing skill and follow this 8-step protocol in this session:

1. Artifact-first scan (MANDATORY before any device_ call).* Glob **/test-app/.rn-agent/actions/*.yaml (and .rn-agent/skeleton.yaml) within the current project AND the sibling workspace at ../rn-dev-agent-workspace/test-app/.rn-agent/actions/. For each candidate, read the file header / appId and decide if it matches the requested feature by:

   • filename keyword overlap with $ARGUMENTS
   • first-comment-block intent overlap with $ARGUMENTS
   • produces: overlap with the goal state (D1209) — an action whose produces includes the state your task requires (e.g. authenticated: true when the task needs an authenticated session) is a useful prologue even if its intent doesn't match the full task. See "Hybrid composition" in CLAUDE-MD-TEMPLATE for the loop. If a full match exists, REPLAY IT FIRST — but call cdp_status once before any replay (the environment gate applies to replays too, not just manual walks; if it fails, route to /rn-dev-agent:setup instead of running a doomed flow). Replay with cdp_run_action({ actionId, params }); the tool preflights the exact runner and action pin before any UI mutation. If the replay passes, you have your evidence — proceed to step 7 (verification + generate-or-refresh artifact). If the replay fails with a concrete error (Element not found, assertion failed), fix the flow rather than abandoning to manual primitives.

   If only a partial match exists (an action whose produces covers part of your task's required state — e.g. login → authenticated, but the rest of the task is novel), use the action as a prologue: replay it via cdp_run_action({ actionId, params }), re-verify state with cdp_navigation_state + cdp_store_state, then continue with steps 1–7 below for the novel part. Save the new action covering the full task at step 7. This is the hybrid-composition path — it's the default, not an escape hatch.

   Falling back to device_* walks WITHOUT having tried existing flows (full or partial) is a captured anti-pattern (see feedback_execute_artifacts_before_manual.md in auto-memory). Run /rn-dev-agent:list-learned-actions if you want to inspect the inventory — the Produces column shows what state each action establishes.

2. Environment check — call cdp_status. If it fails, stop and tell the user to run /rn-dev-agent:setup.

3. Understand the feature — read implementation files, find testIDs, routes, store slices.

4. Plan the test — write test steps and expected outcomes BEFORE executing.

5. Navigate to start — use cdp_navigate or device_deeplink to reach the starting screen.

6. Execute and verify — for each step:

   • Act (device_press, device_fill, device_find)
   • Wait (assertVisible or 1-2s settle)
   • Verify UI (cdp_component_tree(filter=...))
   • Verify data (cdp_store_state(path=...) + cdp_network_log)

7. Edge cases — test empty state, error state, back navigation, rapid taps.

8. Generate or refresh persistent test (MANDATORY). Always end the run by ensuring a Maestro flow exists for the tested feature:

   • If step 0 found and replayed an existing flow: re-validate it still covers the new edge cases; if you discovered a gap, ADD steps to the existing flow (don't fork a new file).
   • If step 0 found NO matching flow: prefer auto-emission over hand- authoring. Wrap the manual walk between cdp_record_test_start and cdp_record_test_stop, then cdp_record_test_save_as_action to write <test-app>/.rn-agent/actions/<feature-slug>.yaml with the metadata header pre-populated (id, intent, tags, mutates, status, enginePin — see skills/rn-testing/SKILL.md "Reusable Action Metadata Schema"). Hand-edit the result to parameterise input strings via ${VAR} placeholders and add a when: visible: id: tab-X self-bootstrap if the flow assumes a starting screen. If <test-app>/.rn-agent/skeleton.yaml exists, add any new testIDs the flow references there too.
   • If step 0 found a flow that doesn't cleanly extend (different feature overlap), still add a NEW flow. Two short flows beat one tangled flow.