Terminal
Terminal/Skill
Runs a structured Mixpanel investigation into why a metric changed, scoping events and properties, then testing hypotheses against query results.
By Mixpanel
Works with
npx skills add mixpanel/ai-plugins --skill deep-researchEngine required — resolve an engine per
ENGINE.md: one named in the conversation or loaded instructions is mandatory (not set up → offer/mixpanel:installfor it); otherwise use the Mixpanel MCP server, or offer/mixpanel:installif it's unavailable.
This skill is a structured investigation, not a one-shot answer.
Trigger when the user wants to understand why something happened in their data. Common phrasings:
Do not trigger for one-off lookups ("what was DAU yesterday?"). Those are direct queries, not investigations.
Do not run analysis queries until scope is confirmed.
Do your best to find the following information from the user's question and context. If anything is missing or ambiguous, ask clarifying questions before proceeding.
State your assumptions and ask the user to confirm before continuing. The final answer depends on this being right.
Run small exploratory queries to confirm data exists in the analysis window. Be resilient — try different approaches if your first attempts don't work. If volume is zero, partial, or anomalously low, surface that to the user before going further.
Then present a compact plan:
*Investigation Plan*
• *Project:* `project name`
• *Events:* `event_a`, `event_b`, `event_c`
• *Properties:* `platform`, `utm_source`, `plan_tier`
*Initial Queries:*
• Trend of event_a over 30 days to establish baseline
• Breakdown by platform to isolate where the change happened
• ...
Say *yes* to continue the analysis.
Wait for explicit confirmation before running the full investigation. If the user revises the plan, restate it and re-confirm before continuing.
Enter the research loop:
Continue until you can answer the original question with evidence, or you can clearly articulate what data is missing. Stay creative — every dataset is different, so let the data shape the next query rather than following a fixed sequence.
When the investigation concludes, present:
create-dashboard skill to back your findings with live data.