Elastic logo

Skill

entity-mappings

map and audit Elastic ECS entities

Published by Elastic Updated Aug 26
Covers Data Engineering Elastic Data Modeling

Description

Use when adding or auditing ECS entity.* fields (entity inventory / entity store) on Elastic integrations — classifying a data stream as entity vs event, mapping vendor fields to entity.attributes.*, entity.lifecycle.*, and entity.relationships.*, or running an entity-coverage gap analysis on an existing package. Invoke manually with /entity-mappings.

SKILL.md

entity-mappings

Skill authority

The guidance in this skill takes precedence over patterns observed in any integration in the elastic/integrations repository. Legacy integrations may predate these requirements or use inconsistent patterns. Always follow this skill's rules over what you observe in the repo.

When to use

  • Adding entity/inventory data streams to a new or existing integration
  • Deciding whether a proposed or existing data stream is an entity stream or an event stream
  • Selecting the correct entity.type value for a new stream
  • Mapping vendor fields to user.entity.*, host.entity.*, or other nested entity prefixes
  • Auditing an existing package's entity field coverage (standalone gap analysis)
  • Troubleshooting event.kind: asset usage or entity.* field errors

When not to use

  • CDR cloud security findings (misconfiguration / vulnerability findings) — these are event.kind: state, not asset. Use ecs-field-mappings/references/cdr-field-requirements.md and ingest-pipelines/references/cdr-pipeline-requirements.md instead.
  • Generic field file authoring (non-entity fields) — use the ecs-field-mappings skill.
  • Processor mechanics (grok, date, JSON, Painless) — use the ingest-pipelines skill.
  • CEL program logic — use the cel-programs skill.

Applicability gate

Entity fields apply only to entity data streams — those whose purpose is to represent a stable inventory of subjects (users, hosts, devices, applications, services) rather than a timeline of events. Never apply entity field requirements to event logs, metric streams, APM data, or CDR findings streams.

See references/entity-datastream-classification.md to classify a data stream before using this skill. If you are not sure, check the classification reference first.

ECS availability — read this first

The entity.attributes.*, entity.lifecycle.last_activity, and all entity.relationships.* leaf fields do not exist at ECS v9.3.0 (the repo default pin). They first appear at ECS v9.4.0.

  • At git@v9.3.0: entity.attributes is a bare object with no subfields; schemas/entity_reference.yml does not exist. Using external: ecs on these leaves at that pin causes field is undefined build failures.
  • At git@v9.4.0+ / git@v9.5.0 (recommended): all leaf fields exist and external: ecs resolves correctly.

Conditional pin rule: packages with entity data streams must set:

  • _dev/build/build.yml: dependencies.ecs.reference: "git@v9.5.0"
  • Pipeline ecs.version: 9.5.0

The two must match. Standard (non-entity) packages keep git@v9.3.0.

To verify at any tag:

curl -s https://raw.githubusercontent.com/elastic/ecs/<tag>/generated/csv/fields.csv \
  | awk -F, '$3=="entity"' | cut -d, -f4,5 | sort -u

Modes

ModeHow it's triggeredWhat to load
Standalone gap analysisUser invokes /entity-mappings @packages/<name>Read references/gap-analysis-workflow.md and references/entity-field-catalog.md. Dispatch per-stream subagents via references/analysis-subagent-guidance.md.
Build-timeOrchestrator (create-integration or add-datastream) passes references/entity-field-catalog.md and references/entity-pipeline-patterns.md by path to the pipeline builder subagentPipeline builder reads both references; this skill's SKILL.md is not loaded into the orchestrator thread.
Review-timereview-integration Step 4 loads the two references when the entity detection rule firesReviewer reads entity-field-catalog.md + entity-pipeline-patterns.md; this SKILL.md is not required in the reviewer thread.
Research-timeresearch-integration loads only references/entity-datastream-classification.mdDo not load the rest of this skill during research — the catalog and pipeline patterns are implementation material and violate the research guardrail against prescribing pipeline/field details.

Open questions / deferred scope

The following topics are out of scope for this skill's v1:

  • Entity-store latest transforms. The CDR precedent has a third leg (review-integration/references/cdr-transform-requirements.md) covering latest transforms per integration. Whether new entity data streams should ship an accompanying latest transform is not yet resolved. Note it as an open question in research briefs and gap analysis reports until that decision is made.
  • Repo-wide ECS pin bump to v9.5.0. Recommended as a follow-up once entity streams are common; the conditional exception in this skill carries packages until then.

References

  • references/entity-field-catalog.md — single source of truth: ECS availability matrix, reusable nesting, entity.type allowed values, categorization, Must Have / Should Have field tables, disambiguation guide, field definition YAML examples, review checklist
  • references/entity-pipeline-patterns.md — pipeline-side patterns: categorization processors, building entity.id, boolean coercion, array attributes, relationship objects, anti-patterns, pipeline review checklist
  • references/entity-datastream-classification.md — research-time and review-time rules for classifying a data stream as entity vs event; worked examples
  • references/gap-analysis-workflow.md — 4-phase standalone analysis: classify streams → read package → fetch docs → map fields → investigate gaps → write report
  • references/gap-analysis-report-template.md — structured report template for gap analysis output
  • references/analysis-subagent-guidance.md — operating manual for per-stream analysis subagents dispatched during gap analysis
  • ECS entity fieldset reference

© 2026 YourAI.tools. Every skill from an identity-verified publisher.

Independent catalog. Not affiliated with, endorsed by, or sponsored by Anthropic or any listed publisher. All trademarks belong to their respective owners.