Skill v1.0.0
currentTrusted Publisher100/100version: "1.0.0" name: analytics description: Use when a Process Mining overview, initial process selection, or an analytical follow-up needs routing. allowed-tools: Read, AskUserQuestion
Process overview and routing
Use the shared contract once per fresh analysis context. Recipes name native functions, not provider-prefixed host tools; bind them to discovered schemas.
Shortest trace
- Reuse the selected process and fresh metadata. Otherwise
get_processes {}->
resolve the returned P (ask on duplicates) -> get_process_details_v2 {processId:P} when metadata is missing. An empty process list stops analysis.
get_process_overall_metrics_v2 {S,filterOptions:F}. Omit absent filters.
Report returned case/event/variant counts and mean CaseDuration/CaseWaitingTime; preserve native units. Missing metrics are not zero. For utilization with zero active time or start-only data, check the zero-duration convention in filters and units; do not infer productive time.
- Only for requested path concentration:
get_variants_with_metrics_v2 {S,filterOptions:F,itemsPerPage:5,itemsToSkip:0,metricToSortBy:"CaseCount",sortOrder:"Descending"}.
Reuse completed evidence; continue necessary pages or checks without fixed analytical-call or total-row quotas. Stop at an answer, not an arbitrary count; do not sweep or repeat unchanged queries. Do not force an overview or investigation loop before a specialist's exact query. Broad business questions or competing explanations use investigate-process: establish only missing outcome definitions that change interpretation, then test the relevant claim.
| Follow-up | Load only this skill | |
|---|---|---|
| Broad business question, observed pattern, competing explanations | investigate-process | |
| Slow activities, waiting, handoffs, cumulative workload | analyze-performance | |
| Common/slow variants, repeated work, a specific case | analyze-variants | |
| Explanatory attributes, association hypotheses | analyze-drivers | |
| Before/after or A/B cohorts | compare-cohorts | |
| Missing metric, saved metric reuse, ephemeral formula | derive-metric | |
| Interacting objects or leading-object executions | analyze-objects |
Stop when answered. Return a brief finding, at most five headline rows, scope/version, metric aggregation/units and coverage; retain compact context for the next question. Views, visualizations and saved metrics cannot be created through these tools.