razbakov/skills/analytics-check

Analytics Check

Query PostHog analytics for DAU (Daily Active Users) across all tracked projects or a single project.

Trigger

• "DAU report" → portfolio-wide report (all projects)
• "analytics for <project>" → single project
• Part of daily/weekly review when analytics check is needed

Configuration

PostHog has two instances. API keys are in ~/.zshrc:

| Instance | Env Var | Host | |----------|---------|------| | EU | POSTHOG_READ_API_KEY | eu.i.posthog.com | | US | POSTHOG_READ_US_API_KEY | us.i.posthog.com |

The project registry (PostHog IDs, instances, URLs) lives in the project's CLAUDE.md under "PostHog Analytics". Read it to get the current list — don't hardcode project IDs in this skill.

PostHog MCP (mcp__posthog__*) is scoped to a single project (WeDance US). For cross-project queries, always use the curl API.

Process: Portfolio DAU Report

1. Load config

source ~/.zshrc 2>/dev/null

Read the PostHog Analytics table from the project CLAUDE.md to get project names, IDs, and instances. Skip projects with - as PostHog ID (no tracking set up).

2. Query all projects

Use the PostHog Query API (/api/projects/<id>/query/) with a TrendsQuery. The legacy /insights/trend/ endpoint is deprecated and returns 403.

curl -s --max-time 20 "https://${instance}.i.posthog.com/api/projects/${project_id}/query/" \
  -H "Authorization: Bearer ${api_key}" \
  -H "Content-Type: application/json" \
  -d '{"query":{"kind":"TrendsQuery","series":[{"event":"$pageview","math":"dau"}],"dateRange":{"date_from":"-7d"}}}'

Response structure: results[0].labels (date strings) and results[0].data (DAU counts).

EU PostHog can be flaky (503 "no healthy upstream"). If a project fails, retry once after a short pause. If it still fails, mark as UNAVAILABLE and continue — don't block the whole report.

Query projects sequentially (not in parallel) to avoid overwhelming the API. Use --max-time 20 to avoid hanging on unresponsive endpoints.

3. Present as summary table

Format all projects in a single table sorted by 7-day average DAU (highest first):

| Project | Apr 2 | Apr 3 | ... | Apr 9 | 7d Avg | |---------|-------|-------|-----|-------|--------| | WeDance v3 | 20 | 21 | ... | 39 | 23 |

4. Signals

After the table, surface signals — not just observations, but routable findings. The point of this step is that a list of numbers is barely useful; what's useful is a short interpretation that someone could act on tomorrow.

• Top performer: which project carries the portfolio (% of total DAU). If one project dominates >50%, that's worth naming — it means a "portfolio" framing isn't yet visible in traffic.
• Anomaly spikes: any single day >2× a project's 7-day baseline. Name the day, name the magnitude, and ask what was published/shared. If the cause is obvious (weekend, launch day, known event), say so.
• Tracking gaps: any project with 0 events for 7 days running, OR fewer active days than expected for a project that's claimed to be live. These are likely SDK/wiring problems, not user behavior. Flag them clearly so they don't get read as "no traffic".
• OKR/target gap: if the project has a stated traffic or DAU target (check the project's CLAUDE.md, OKRs, README, or the user's stated goals), compute the gap. "X DAU vs target Y = Zx short." Skip this bullet if there's no stated target.
• Unavailable projects (API failures): list separately so they're not confused with zero-traffic.

5. Action items (only if invoked from a review or with "save")

End the report with a small action list — one bullet per non-trivial signal, with a clear owner. Owners come from the project/org context (read the project's CLAUDE.md for the agent team or roles); don't hardcode names. Typical mappings:

• Tracking gaps → engineering / CTO role
• Content or traffic spikes worth replicating → content / growth role
• Event-driven spikes → community / events role
• Target-vs-reality gaps → strategy / business role

If the project has no agent team defined, leave the owner as "TBD" or address it to the user.

6. Save snapshot (optional)

If the user said "save", or this skill was invoked from a daily/weekly review, persist the full report so other agents can reference it later. Read the project's CLAUDE.md to find the conventional sessions/snapshots directory (commonly ops/sessions/ or sessions/). Filename:

<sessions-dir>/YYYY-MM-DD-portfolio-dau-snapshot.md

Use today's date. Include the per-day matrix, per-project averages, signals, and action items.

Process: Single Project

When the user asks about a specific project:

1. Look up the project in the CLAUDE.md PostHog table
2. Query just that project using the same curl approach
3. Present the 7-day DAU table
4. Calculate 7-day average (exclude partial today if it's early in the day)
5. Note trend direction (up/down/stable)
6. If PostHog MCP is available for that project, also check test account filters and link to the insight

Troubleshooting

• 503 / "no healthy upstream": PostHog instance is having issues. Retry once, then mark unavailable.
• "Legacy insight endpoints are not available": You're using the old /insights/trend/ endpoint. Switch to /query/ with TrendsQuery.
• Empty JSON response: The curl request timed out silently. Increase --max-time.
• All zeros for a project: Either no traffic, tracking not installed, or the PostHog project ID is wrong. Suggest the user verify the tracking snippet is on the live site.