replay-session

name	replay-session
description	Read a Tracey session dump file, map events to code paths, reconstruct the user journey, and identify where things went wrong. Use when user provides a crash dump, says "replay this session", "what happened before the crash", "analyze this dump", "tracey dump", or provides a path to a replay JSON file.

Session Dump

$ARGUMENTS

Process

1. Read and parse the Tracey JSON replay file at the given path. Extract: session ID, device info, crash flag, event timeline, stacktrace (if crash).

2. Build timeline table — for each event, resolve corresponding source:

   #	Time	Type	Detail	Code Path

   Event type mapping:

   • SCREEN → destination in navigation/.../NavGraph.kt
   • CLICK/SWIPE/SCROLL/LONG_PRESS/PINCH → Screen composable at active route
   • LOG → Tracey.log() call in ViewModel (Grep for the log message)
   • FOREGROUND/BACKGROUND → lifecycle event
   • CRASH → stacktrace + AppError mapping

3. Identify failure point:

   • Last successful state/event before crash
   • First error event in sequence
   • Time gap between last gesture and crash (user-perceived latency)
   • If no crash — identify anomalous state transitions

4. If captureAndExportTest() output is available alongside the dump:

   • Show the generated Kotlin test code
   • Adapt to project conventions (AAA pattern, test naming should X when Y)
   • Suggest target test file per testing.md rules

5. If claude-in-mobile MCP is available and user wants replay:

   • mcp__claude-in-mobile__launch_app(package="com.portfolio.nutrisport.debug")
   • For each gesture event: mcp__claude-in-mobile__tap(x, y) + mcp__claude-in-mobile__wait(ms=delta)
   • mcp__claude-in-mobile__screenshot() at failure point
   • Compare with expected state from dump

6. Generate markdown report:

   Session Replay: {sessionId}

   Platform: {platform} | Duration: {first event} → {last event} Screens visited: {list} Crash: {yes/no} | Exception: {type at file:line}

   Timeline

   (table from step 2)

   Failure Analysis

   • Last good state: {event before failure}
   • First error: {error event}
   • Root cause: {analysis}

   Recommended Next Steps

   • /debug-crash {dump_path} — for full fix workflow
   • /gen-test — generate regression test from dump

Rules

• Always read the JSON file first — never guess from filename
• Map every event to actual code paths in the project
• If dump is not a crash (manual capture) — analyze for anomalies
• Keep report concise — focus on the failure, not every event