Finding Replay For Issue

Finding the best replay for an error tracking issue

When a user says “show me a replay for this error” or “find a recording for issue X”, the goal isn’t just any linked session — it’s the one that best shows what led to the error. Popular issues can have hundreds of linked sessions, and most are crash-only fragments or duplicate occurrences. This skill picks the most useful one.

Available tools

Workflow

Step 1 — Get the issue details

Fetch the error tracking issue to understand what you’re looking for:

posthog:query-error-tracking-issue
{
  "issueId": "<issue_id>"
}

Note the issue’s fingerprint, name, and description — you’ll need the fingerprint to find linked sessions.

Step 2 — Find sessions with this error

Query exception events to get session IDs where this error occurred. Order by recency and include basic context:

posthog:execute-sql
SELECT
    $session_id AS session_id,
    count() AS occurrences,
    min(timestamp) AS first_seen,
    max(timestamp) AS last_seen,
    any(properties.$current_url) AS url
FROM events
WHERE event = '$exception'
    AND properties.$exception_fingerprint = '<fingerprint>'
    AND $session_id IS NOT NULL
    AND timestamp > now() - INTERVAL 30 DAY
GROUP BY session_id
ORDER BY last_seen DESC
LIMIT 20

This gives you up to 20 candidate sessions. More candidates means better selection.

Step 3 — Rank the candidates

Fetch recording metadata for the candidate sessions to rank them:

posthog:query-session-recordings-list
{
  "session_ids": ["<id1>", "<id2>", "<id3>", ...],
  "date_from": "-30d"
}

Pick the best recording by filtering out bad candidates, then ranking what’s left:

Filter out:

• Sessions under 10 seconds (crash-only fragments, no pre-error context)
• Sessions over 1 hour (too much data to load, error is a needle in a haystack)

Rank by:

1. Sweet-spot duration — 2-15 minutes is ideal. Long enough to show the user’s journey before the error, short enough to be practical to watch or summarize.
2. Active time ratio — compare active_seconds to recording_duration. A 20-minute recording with 10 seconds of activity is mostly idle tabs — the user walked away. Prefer sessions where active_seconds / recording_duration is above 0.3 (30%).
3. Activity score — higher activity_score means the user was actively interacting, not idle. More interesting to watch.
4. Recency — more recent sessions reflect current app behavior.

Step 4 — Present the finding

Fetch full details for the selected recording:

posthog:session-recording-get
{
  "id": "<best_recording_id>"
}

Present to the user:

• The recording with a link to watch it
• Why this one — briefly explain the selection (“longest session with the error, user was browsing 3 pages before hitting it”)
• Pre-error context — what pages the user visited and key actions before the exception, derived from the events query in step 2 (the url and first_seen columns)
• Error frequency — how many times the error occurred in this session

Optional: AI summary via Replay Vision

If the user wants a narrative summary without watching, use Replay Vision — “check-then-scan”, since a scanner can only observe a given session once.

1. Check for an existing summary on the selected recording:

   jsonCopy

   posthog:vision-observations-list
   {
     "session_id": "<best_recording_id>"
   }

   If an observation has scanner_snapshot.scanner_type summarizer and status succeeded, read scanner_result.model_output (title, summary, intent, outcome, friction_points, keywords) — done.

2. Find a summarizer scanner if none exists:

   jsonCopy

   posthog:vision-scanners-list
   {
     "scanner_type": "summarizer"
   }

   One → use it. More than one → ask the user which (show name + prompt). None → offer to create one via the creating-replay-vision-scanners skill.

3. Scan the recording with the chosen scanner (async, several minutes):

   jsonCopy

   posthog:vision-scanners-scan-session
   {
     "id": "<scanner_id>",
     "session_id": "<best_recording_id>"
   }

4. Retrieve by polling vision-observations-list until succeeded.

Tips

• If all candidate sessions are very short (<10 seconds), the error likely crashes the page immediately. Note this — it’s useful context even without a long replay.
• When the issue has very few linked sessions (<3), skip the ranking and just present what’s available with a note about the small sample.
• If $session_id is null on many exception events, session replay may not be enabled for the affected users. Mention this as a possible gap.
• Replay Vision has no per-call focus parameter — a summarizer scanner’s focus comes from its own prompt. For error-focused summaries, prefer (or create) a summarizer scanner whose prompt targets error/exception context rather than the whole session.

Related skills

• investigating-error-issue — the quantitative side: volume, breakdowns, and stack traces for the same issue
• investigating-replay — deep-dive one of the linked sessions
• diagnosing-missing-recordings — when exception events have no $session_id or the linked recordings are gone