# data

> Make product and business data trustworthy, governed, usable. Installs as an agent skill.

Ask an agent to "track this event" or "add a data pipeline" without this skill and it ships an
ad-hoc payload with no naming grammar, no schema, no test — a warehouse table nobody trusts and
nobody can safely change. `data` is a faceted agent skill for the **analytical and governed data
plane**: event tracking and contracts, ingestion/movement, transformation and modeling, the
metrics/semantic layer and BI, data quality and testing, lineage and observability,
governance/PII/retention, experiment measurement, and platform/cost. Every request routes through
one primary job (of ten) composed with at most one base surface (of four), plus the additive
`surface-regulated` overlay when GDPR/CCPA/HIPAA/SOC2 applies — never the whole reference library
at once. Trust is a gate, not a dashboard: the convergent DQ taxonomy (9 core buckets + PR-diff +
PII), a PK-uniqueness floor on every model, and a PR-time blast-radius diff before merge. Every
tool version, warehouse price, and license is re-verified against the live source and date-stamped
— read the LICENSE file, never a repo's API license field. Claude Code is the reference host; it's
plain Agent Skills format (agentskills.io), so it installs on Codex, Cursor, Antigravity, opencode,
Grok Build, and Hermes too.

## Install

- npm/skills.sh: `npx skills add gabros20/data-skill -g -y`
- Clone + installer (per-host targets: claude | codex | cursor | antigravity | opencode | grok | hermes | agents | all): `git clone https://github.com/gabros20/data-skill && cd data-skill && ./install.sh codex`
- Manual copy: `cp -R skills/data ~/.claude/skills/data`

## Docs

- [README.md](https://raw.githubusercontent.com/gabros20/data-skill/main/README.md): overview, install, workflow, and repository map.
- [skills/data/SKILL.md](https://raw.githubusercontent.com/gabros20/data-skill/main/skills/data/SKILL.md): runtime router, workflow, artifact contract, and completion rules.
- [docs/installation.md](https://raw.githubusercontent.com/gabros20/data-skill/main/docs/installation.md): installation and verification by client.
- [docs/usage.md](https://raw.githubusercontent.com/gabros20/data-skill/main/docs/usage.md): activation boundary, routes, outputs, and completion.
- [docs/recipes.md](https://raw.githubusercontent.com/gabros20/data-skill/main/docs/recipes.md): representative starting prompts.

## Reference inventory (`skills/data/references/`)

Primary jobs (ten):
- `event-taxonomy-and-tracking.md` — naming grammar, tracking plan, instrumentation QA, consent-aware collection.
- `data-contracts-and-schema-evolution.md` — ODCS contract shape, 7 compatibility modes, write-audit-publish.
- `ingestion-and-movement.md` — sync-strategy decision table, CDC + outbox, destructive-action gating.
- `transformation-and-modeling.md` — narrow-the-DAG/widen-the-tables, materialization ladder, Kimball/OBT/Vault.
- `metrics-semantics-and-bi.md` — one-authoritative-definition semantic layer, code-first vs GUI BI, NL-to-data.
- `quality-and-testing.md` ⭐ — the convergent DQ taxonomy, known/unknown unknowns, PK floor, PR-time blast-radius diff.
- `lineage-and-observability.md` — OpenLineage model, five pillars/data downtime, catalog choice.
- `governance-pii-and-retention.md` — platform-layer masking, Art.17-with-exceptions erasure, retention.
- `experiment-measurement-foundations.md` — SRM-before-trust, deterministic salted assignment, CUPED.
- `platform-and-cost.md` — small-data-first (COST rule), orchestrator choice, the licensing map.

Surface overlays (five):
- `surface-product-analytics.md` — default/start-simplest; tracking plan as spine.
- `surface-warehouse-platform.md` — central platform; ELT + semantic layer + BI serving.
- `surface-operational-activation.md` — CDC/reverse-ETL, freshness SLAs, idempotency.
- `surface-ai-data-plane.md` — RAG corpus as governed ETL, point-in-time correctness, eval datasets.
- `surface-regulated.md` ⭐ (additive) — GDPR/CCPA/HIPAA/SOC2 consent gating, masking, erasure ops.

Pipeline:
- `handoff.md` — standalone vs pipeline behavior and the `handoff.yaml` companion (with `volatile_facts`).

## Optional

- [Visual guide](https://data-skill.vercel.app): the router, the quality-and-testing flagship, the RAG three-way seam, and the reference map on one page.