Back to Claude Mem

PostHog Self-driving Setup Report

posthog-self-driving-report.md

13.15.38.3 KB
Original Source

PostHog Self-driving Setup Report

Generated 2026-07-14 for project CMEM (ID 463218)

Summary

PostHog Self-driving was configured for the claude-mem project. Error tracking (three trigger types), session replay, GitHub Issues, and support ticket signal sources were enabled; the scout troop was tuned to three scouts matching this project's primary surfaces (AI/LLM usage and MCP tool calls). Findings will start appearing in the Self-driving inbox within ~30 minutes.


AI data processing

Approved. Organization-level AI data processing consent was granted before this run.


GitHub

Already connected — GitHub App integration (thedotmack, integration ID 175967) was present at setup time, connected on 2026-06-10. One repository is synced: thedotmack/claude-mem.


Products enabled

ProductStatusNotes
Session ReplayAlready activeRecordings flowing from cmem.ai; products-enable API not available in this deployment — server state confirmed by live recordings
Error TrackingAlready active1.1M+ events via posthog-node with enableExceptionAutocapture; products-enable API not available
Support (Conversations)Follow-up requiredProduct could not be enabled via API (tool unavailable); see follow-ups for manual step

This project uses posthog-node (backend), not posthog-js. No posthog.init override check was needed. Session replay for the web front-end (cmem.ai) appears to be instrumented from a separate repo or snippet.


Signal sources

source_productsource_typeAction
error_trackingissue_createdEnabled (new, ID 019f6318-32fa-73a9-914f-55617f6800de)
error_trackingissue_reopenedEnabled (new, ID 019f6318-3775-7090-a2e5-d98f893734d0)
error_trackingissue_spikingEnabled (new, ID 019f6318-39b6-7161-8bce-8e03a206d92b)
session_replaysession_analysis_clusterEnabled (new, ID 019f6318-3dd1-7dae-b4a0-36eba57cb746; default 10% sample rate)
conversationsticketEnabled (new, ID 019f6318-4047-764c-94fc-ee62a511a9b2; dormant until a channel is connected)
signals_scoutcross_source_issueOn by default — no config row needed; scout findings reach the inbox automatically
llm_analyticsevaluation_reportSkipped — internal-only responder, not user-facing
logsSkipped — not a v1 responder

Connected tools

ToolStatus
GitHub IssuesAlready connectedGithub warehouse source (ID 019eb017-8f11-0000-8ce2-d7bed20cb960) was present; 1,766 issues synced as of 2026-07-14. Responder (github / issue, ID 019f6324-8a46-71ab-ad0b-8cea5b4fb795) enabled.
LinearNot used — not selected
ZendeskNot used — not selected
pganalyzeNot used — not selected

Note: the GitHub warehouse source has a token error on the stargazers table (Access forbidden). The issues table (the one the responder reads) is syncing cleanly. See follow-ups.


Scout troop

3 enabled, 23 disabled.

Enabled

ScoutReason
signals-scout-generalAlways on — sweeps cross-product correlations and surfaces no specialist covers
signals-scout-ai-observabilityThis product wraps Claude/Anthropic APIs — LLM cost, latency, and error regressions are directly on-product
signals-scout-mcp-tool-callsThis product IS an MCP server — $mcp_tool_call telemetry covers tool failure rates and confusing schemas

Disabled

ScoutReason
signals-scout-error-trackingCovered by the native error_tracking source (all three trigger types enabled above)
signals-scout-session-replayCovered by the native session_replay source enabled above
signals-scout-product-analyticsNo confirmed active funnels/retention insights — enable if you build out product analytics dashboards
signals-scout-web-analyticsNo confirmed UTM/referrer tracking in this repo — enable if you add web analytics to cmem.ai
signals-scout-feature-flagsFeature flag usage not confirmed — enable if you adopt flags
signals-scout-surveys0 surveys in use
signals-scout-revenue-analyticsNo payment SDK found
signals-scout-logsPostHog logs product not confirmed in use
signals-scout-csp-violationsNo CSP reporting configured
signals-scout-experimentsNo active A/B experiments
signals-scout-customer-analyticsNo group/accounts analytics confirmed
signals-scout-data-pipelinesNo CDP destinations or hog flows found
signals-scout-replay-visionReplay Vision scanners not configured
signals-scout-apmNo OpenTelemetry/APM tracing found
signals-scout-anomaly-detectionCovered adequately by general + specialists for this project size
signals-scout-observability-gapsCan enable later if event coverage gaps become a concern
signals-scout-health-checksCan enable if setup health becomes a concern
signals-scout-inbox-validationNot useful on a fresh setup — no shipped fixes to validate yet
signals-scout-ingestion-warningsCan enable if ingestion errors appear
signals-scout-insight-alertsNo configured insight alerts found
signals-scout-skills-storeInternal PostHog skill hygiene scout
signals-scout-data-warehouseGitHub source token issue on stargazers — see follow-ups
signals-scout-web-vitalsNo $web_vitals events confirmed from this repo

Custom scouts

Two surfaces were proposed and declined:

Proposed scoutSurfaceWhy declined
Memory pipeline healthobserver_turn_rollupcontext_injected_rollup volume/ratio degradationUser dismissed the proposal
Install-to-first-memory funnelinstall_completedworker_started → first observationUser dismissed the proposal

Gap analysis ruled out:

  • Chroma dependency health — already covered by the error_tracking native source (ChromaUnavailableError is the dominant issue)

Noise escape hatch: to put any scout into dry-run (it runs and logs but emits nothing to the inbox), set emit: false on its config in PostHog > Self-driving settings.


Follow-ups

  • Enable Support/Conversations productproducts-enable API was not available in this PostHog deployment. Enable Session Replay, Error Tracking, and Conversations manually in Project settings > Products if not already on.
  • Connect a Support inbound channel — The conversations/ticket signal source is enabled and waiting. Connect an email, inbox, or Slack channel in PostHog so support tickets start reaching the inbox. (Integrations settings)
  • Fix GitHub token permissions — The stargazers table is failing with "Access forbidden / rate limits". Check the GitHub App token scope at Integrations settings — the issues table syncs cleanly so this doesn't block the Self-driving responder, but it may affect other warehouse queries.
  • Consider enabling signals-scout-memory-pipeline — A custom scout watching observer_turn_rollup / context_injected_rollup volume and ratio degradation would catch silent pipeline failures before users notice worse memory quality. Create it via Self-driving scouts if you want this coverage.
  • Consider enabling signals-scout-install-funnel — A custom scout watching install_completedworker_started → first observation would catch broken first-run flows and platform-specific regressions. Create it if install conversion becomes a priority metric.

What happens next

The scout coordinator picks up fresh configs within ~30 minutes. Scouts run daily (1,440-minute interval) and emit findings as reports into the Self-driving inbox. Error tracking and session replay signals arrive sooner — they trigger on individual events as they land. Immediately-actionable reports can be turned into coding tasks directly from the inbox.