error-forensics

Error Forensics

Find what's breaking in production — JavaScript errors, network failures, console exceptions — and trace them to root cause using real session evidence.

When to Use

• "Are users hitting JavaScript errors on the settings page?"
• "Find sessions with 504 errors on /api/orders"
• "What's the most common console error right now?"
• "A user reported the checkout broke — find their session and tell me what went wrong"
• "Is the TypeError on /dashboard affecting many users or just one?"
• "What errors started appearing after the last deploy?"

Workflow

Step 1: Discover error patterns

Start broad. Call fullstory:get_opportunities to surface error-related frustration signals — these will include console errors, network failures, and renderer errors ranked by user count.

If you're investigating a specific error or page, build a metric instead:

fullstory:build_metric(
  query="console errors on /settings",
  output_type="top_n"
)

Step 2: Quantify

Call fullstory:get_opportunity_stats or fullstory:compute_metric to get:

• How many users are affected
• Device/browser breakdown (is it Safari-only? Mobile-only?)
• Page concentration (which pages is it happening on?)
• Trend (is it getting worse? Did it start after a deploy date?)
• Rate vs site average (is this page abnormally error-prone?)

Step 3: Find sessions

Call fullstory:get_sessions_for_opportunity or fullstory:get_sessions(metric_id) to get 3-5 sessions with the error.

Step 4: Investigate each session

For each session, use the session-context agent with a focused task:

"Was there a JavaScript error in this session? 
If yes: what was the error message, what page was the user on, 
what action triggered it, and what was the stack trace?"

Step 5: Open the worst sessions

For sessions that look like the canonical case, use session-review to:

• Open the session: fullstory:session_open(session_url)
• Navigate to the moment of the error
• Use fullstory:session_view to see the UI state at error time
• Use fullstory:session_diff to compare just before/after the error — what changed (or didn't)?