schematize
Skills · Claude Code · Agent Skills

Data is a contract — no negotiation.

Data as a contract, not a side effect: a breaking change requires a new version, delivery via Outbox, safe idempotency and replay, bad data goes to quarantine and deny-by-default on access.

v0.1.0Disciplina · Dadosgithub.com/schematizeme/skill-data

Onderhouden door Lucas Andrade

schematize-data · v0.1.0bash
$ schematize install data
GitHub →download skill (.zip)/data-help · github.com/schematizeme/schematize-cli

Non-negotiable floors

Never switched off "to go faster". Full list in the skill.

Data is a contract

Every piece of data that crosses a boundary has a versioned schema. Removing/renaming/changing a type is a break: it requires a v2 and a documented coexistence window.

Delivery via Outbox

Transactional Outbox (or CDC from the database log); dual-write is vetoed. Every consumer has a DLQ; retry is bounded with backoff, never infinite.

Safe idempotency and replay

Idempotent consumer (running twice doesn't corrupt); deterministic replay rebuilds the same state; side effects deduped by idempotency-key.

Bad data doesn't contaminate downstream

Validation at the edge; whatever fails goes to quarantine/DLQ with reason and origin — never enters the good dataset, never vanishes silently.

Reversible expand-contract migration

expand → migrate → contract, each step with a tested down. A destructive DROP/rename in a single step is vetoed. Timestamps in UTC, IDs UUIDv7/ULID.

Deny-by-default + governed PII

Access denied by default, per tenant server-side; parameterized SQL. PII only with legal basis + retention, out of logs/URLs. Traceable lineage; always archive.

Commands

Installed in .claude/commands/. Run /data-help anytime.

/data-helpLists every command.
/data-contractDefines/audits the data contract: versioned schema, compatibility (back/forward/full), registry, evolution policy; blocks a breaking change without a new version.
/data-pipelineDesigns/audits the pipeline (ETL/ELT batch, streaming, CDC/event sourcing): idempotency, delivery via Outbox, replay, backpressure, DLQ.
/data-qualityBuilds/audits the quality gate: validation at the edge, data tests/expectations, quarantine and dead-letter, data SLA.
/data-loadForce-loads the whole normative body: contracts, pipelines, quality, storage, governance.
/data-claudeCreates or merges the always-on data CLAUDE.md at the repo root.
/data-ccContext compact: writes the handoff to the archive and runs /compact.
/data-handoffWrites the handoff (context.md + checklist.md) without compacting.

For AI agents

Install with no human in the loop. Metadata in llms.txt.

  • Run: git clone https://github.com/schematizeme/skill-data.git && ./skill-data/install.sh
  • Or download: https://github.com/schematizeme/skill-data/releases/latest/download/skill-data.zip and unzip into .claude/skills/
  • Copy assets/CLAUDE.md to the repo root (always-on rule).
  • Confirm with /data-help.

Wil je niet installeren?

Lees de volledige standaarden en kopieer ze rechtstreeks naar je project — of installeer de skill zodat de agent ze voor je afdwingt.

GitHub →Packs