Time Series Data Quality skills for AI agents

Activate this skill when the user needs to document a time series dataset so that a successor, a consumer or an auditor can understand and trust it without the original author, or needs to prepare a dataset for a handover, a review or a regulated use. Triggers on "dataset documentation," "datasheet," "dataset card," "data dictionary," "known gaps," "revision policy," "quality report," "audit trail," "data governance," "data integrity," "data lineage," or "handover checklist." Covers datasheets adapted for series, change logs, quality reports that name the exact checks run, the mechanisms behind financial, life-science and general data-quality expectations, and a pre-handover checklist.

Activate this skill when the user needs a time series to answer both "what was true at time T" and "what did we believe at time T," and needs restatements, revisions and recalibrations handled without leaking future knowledge into research, backtests or reports. Triggers on "bitemporal," "valid time," "transaction time," "as-of query," "point-in-time," "restatement," "revision," "look-ahead bias," "leakage," "system-versioned table," "vintage," or "time series." Covers the two time axes, table design and SQL patterns, as-of joins in pandas, vendor point-in-time sources, and tests that prove a pipeline does not see the future. Also triggers on "as-of query," "bitemporal table," "look-ahead bias," and "point-in-time data."

Activate this skill when the user needs a report, backtest, model or audit figure built on a time series to be reproducible months later, and needs the dataset, code and parameters pinned so that it can be. Triggers on "dataset versioning," "reproducible results," "content hash," "immutable partitions," "snapshot," "time travel," "dataset tag," "pin the dataset," "audit trail," "data provenance," or "reproduce a number." Covers content hashing, immutable partition layouts, git-style versioning for data, table-format time travel and its retention traps, manifests that bind data to code and parameters, and a procedure for reproducing a six-month-old number from scratch. Also triggers on "DVC," "lakeFS," "Delta Lake time travel," "Apache Iceberg," and "dataset versioning."

Activate this skill when the user needs to know, before a consumer does, that a time series is late, short, or shaped differently from yesterday, and needs monitors that alert on data-quality metrics rather than on the values themselves. Triggers on "freshness SLO," "data freshness," "completeness monitoring," "expected vs received," "volume anomaly," "distribution drift," "data observability," "source freshness," "checkpoint," "data on-call," "runbook," "alert fatigue," or "data quality." Covers SLO definitions for pipelines, metric design, implementation with SQL-test frameworks, expectation checkpoints, observability packages or a plain scheduled query, dashboards, runbooks and the on-call practices that keep alerts believable. Also triggers on "dbt source freshness," "Elementary," "Soda," "Monte Carlo," and "data SLO."

Activate this skill when the user needs to find missing points, duplicated or conflicting records, out-of-order arrivals, spikes, flatlines, stale values or silent unit changes in a time series, and to decide which of them to flag rather than fix. Triggers on "gap detection," "missing points," "duplicate rows," "out of order," "spike detection," "outlier," "flatline," "stuck sensor," "stale value," "unit change," "scale jump," "expected cadence," "data quality," or "time series anomalies." Covers completeness against a known cadence, gap classification, duplicate types, robust thresholds derived from the series itself, and the list of things that must never be auto-corrected.

Activate this skill when the user needs to turn quality checks on a time series into a score consumers can act on, hold suspect points aside without losing them, repair gaps and corrections without destroying provenance, and rerun pipelines safely. Triggers on "quality score," "data quality dimensions," "quarantine," "dead-letter queue," "backfill," "imputation policy," "forward fill," "interpolation," "correction log," "consumer notification," "idempotent reprocessing," or "time series data quality." Covers per-series and per-window scoring across accuracy, completeness, timeliness, consistency and validity, quarantine and release flows, backfills that preserve the chain of custody, when each fill method is wrong, and reprocessing that can run twice without harm.

Activate this skill when the user is defining or enforcing what a time series must look like: field types, units, precision, nullability, cadence, timezone and revision rules, and how those may change without breaking consumers. Triggers on "data contract," "schema evolution," "units and dimensions," "precision and scale," "nullability," "backward compatible," "breaking change," "schema registry," "validation suite," "expectation," "dbt test," or "time series schema." Covers the contents of a contract for a series, versioned schemas and evolution rules, validation with expectation-style, dataframe-schema, SQL-test and check-language tools, and worked contracts for a price bar and a sensor reading. Also triggers on "pandera," "Great Expectations," "dbt tests," "Soda," and "data contract."

Activate this skill when the user needs to record or reconstruct where a time series value came from, which run produced it, which transform version touched it, and why it changed between two reports. Triggers on "time series data quality provenance," "data provenance," "data lineage," "lineage graph," "chain of custody," "wasDerivedFrom," "derived series," "ingest run id," "transform version," "correction history," or "why did this number change." Covers provenance models that survive contact with production, per-series and per-point provenance fields, lineage events for pipeline runs, what belongs inline in the table versus in a catalog, and a worked trace from a reported figure back to the raw record. Also triggers on "OpenLineage," "W3C PROV," "Marquez," and "lineage graph."

Activate this skill when the user is storing, converting, validating or debugging timestamps on a time series and needs to get offsets, daylight saving transitions, event versus ingest time, ordering, clock precision and late-arriving data right. Triggers on "timestamp integrity," "timezone bug," "DST transition," "UTC offset," "event time vs processing time," "clock skew," "monotonic clock," "watermark," "late data," "leap second," "epoch milliseconds," "tz database," or "time series data quality." Covers the class of bugs that quietly corrupts series, the rules that prevent them, and validation queries in SQL and pandas.