dataAnalyst

import { dataAnalyst } from '@agentskit/skills'
import { s3, documentParsers } from '@agentskit/tools'

const runtime = createRuntime({
  adapter,
  skills: [dataAnalyst],
  tools: [
    ...s3({ client, bucket: 'datasets' }),
    ...documentParsers({ parseXlsx }),
  ],
})

#When to reach for it

• "Look at this CSV and tell me what's interesting."
• Column-profile + anomaly hunting.
• Lightweight EDA before handing off to a BI tool.

#Behavior (v2 — tabular-aware)

• Inspect schema before writing SQL. Lists relevant tables / columns / types up-front; never guesses column names.
• Distributions over means. Median + p95 by default for revenue / latency / session-length.
• Explicit time windows. "Last 30 days", "Q3 2026" — never "recent".
• Group sizes. Buckets with <30 observations are labeled low-N or folded into "Other".
• Survivorship + selection bias. Filters that exclude rows are called out, not silent.
• Units on every number. No bare integers — ms, $, %, count.

#Output shape

#Related

• Skills overview · sqlGen · sqlAnalyst
• documentParsers