Usage Insights

Usage Insights Skill

This skill helps an organization admin understand how their org uses Beau, by reporting over the platform's append-only analytics event log. It produces periodic, evidence-based summaries: student engagement, content authoring (including human-vs-AI), and commerce (enrollments + revenue), plus answers to ad-hoc questions.

It is read-only and org-scoped: every tool only ever sees the caller's own organization. All tools require the read:analytics permission (admin-only) — teachers and students cannot use them.

Instructions for Claude

0. Preflight check

Confirm access by calling get_engagement_report with no arguments (defaults to the last 30 days).

• If it returns a tool error mentioning authentication, tell the user: "The Beau MCP server is not connected — please check your MCP connection and authenticate."
• If it returns a 403 / permission error, tell the user: "Your account doesn't have the read:analytics permission. An org admin must grant it (it's admin-only)." Then stop.

1. Determine scope

Ask the user (unless they already said):

• Report type — Engagement, Content, Commerce, an Overview (all three), or a specific ad-hoc question.
• Time window — e.g. last 7/30/90 days, a specific month, or a custom range. Convert to ISO dates (from/to). Default is the trailing 30 days.

2. Pull the data

Use the curated report tools for standard reports (one call each, pass from/to):

For ad-hoc questions, use aggregate_events:

• groupBy ∈ event | actorType | subjectType | day | week | month
• metric ∈ count | sum | avg (sum/avg need a field ∈ amountCents | platformFeeCents | score | durationSec)
• optional events (names to filter), from, to
• Examples: lessons completed per week → {events:['lesson.completed'], groupBy:'week', metric:'count'}; revenue per month → {events:['payment.completed'], groupBy:'month', metric:'sum', field:'amountCents'}.

For drill-down (inspect individual rows after an aggregate looks surprising), use query_events (filter by events, from/to, actorId, subjectType, subjectId; paginate with limit/offset).

3. Interpret correctly

• Money is in cents — divide by 100 and show the currency.
• Completion rate = completed ÷ started; call out low rates.
• Test runs are already excluded from engagement (teacher previews don't count as student activity).
• Human vs AI authoring — resourcesByAgent / aiAuthored are content made via the AI assistant or MCP; resourcesByHuman are dashboard edits.
• Empty windows are normal for new orgs — say so rather than implying a problem.

4. Produce the report

Write a concise markdown report:

• A short headline summary (2–3 sentences).
• One section per requested area with the key numbers and a trend note (use the daily/weekly series).
• 2–4 concrete observations or recommendations grounded in the numbers (e.g. "37% of started lessons go uncompleted — consider shorter lessons or a nudge").
• Note the window and that figures are for this organization only.

Do not invent numbers or fill gaps — if a tool returned zero or null, report it as such.

Event reference (for aggregate_events / query_events)

The log records these event families (org-scoped). Names are stable:

• Learning: lesson.started, lesson.completed, feedback.submitted, contact_teacher.triggered
• Authoring (CRUD): resource.created/updated/deleted, bot.*, course.*, image.*, quiz.created, course.resource_added, resource.snapshot_saved, resource.exported, ai_assistant.resource_authored
• Enrollment/commerce: enrollment.created/paused/resumed/completed, payment.initiated/completed/refunded
• Identity/org: user.invited/invitation_accepted/login, consent.submitted/verified, org.settings_changed/stripe_onboarded
• Sharing: share_link.created/consumed/revoked
• Comms: notification.sent (every outbound email)

Useful properties for sum/avg: amountCents, platformFeeCents (payments), score, durationSec (lesson.completed).