parse-grid-failure

Parse Grid Failure

Turn a noisy cloud-grid failure into something a QA engineer can triage in under 30 seconds. Input is $ARGUMENTS — either a session UUID or a JSON blob from BrowserStack/SauceLabs/LambdaTest API.

Requires browserstack MCP for live session fetch — see qa-mobile-pack/mcp/README.md. If MCP is not available, ask the user to paste the session JSON.

Procedure

1. Detect input type:
   • 36-char UUID matching [0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12} → session ID. If browserstack MCP is available, fetch session details. Otherwise stop and request the JSON.
   • Otherwise treat input as raw JSON (BrowserStack automate/sessions/{id}.json shape, SauceLabs jobs/{id} shape, or LambdaTest sessions/{id} shape).
2. Extract the following fields (provider field names vary, use closest match):
   • device (model name, e.g. iPhone 15 Pro)
   • os + os_version
   • locale (default en-US if missing)
   • browser_version or app_version
   • created_at / timestamp
   • failed_step (last step name before failure, or reason)
   • error_message — first 200 chars only
   • video_url / screencast_url
   • logs_url / appium_logs_url
   • network_url / har_url
3. Render the summary using the template below.
4. Generate exactly 3 root-cause hypotheses, each tagged with one of: timing, network, selector, device, data. Base the tags on signals from the error message (e.g. ElementNotInteractable → timing; ECONNRESET/5xx → network; NoSuchElement → selector; OutOfMemory/ANR → device).

Output template

**Failure summary**
- Device: <model> (<os> <os_version>, locale <locale>)
- App/Browser: <version>
- When: <timestamp>
- Failed step: <step name>
- Error: <first 200 chars>

**Artefacts**
- Video: <video_url>
- Logs: <logs_url>
- Network HAR: <network_url>

**Likely causes**
1. [<tag>] <hypothesis>
2. [<tag>] <hypothesis>
3. [<tag>] <hypothesis>