| name | eventcatalog-documentation |
| description | Generates EventCatalog documentation files (services, agents, events, commands, queries, domains, flows, channels, containers, ADRs, data products, entities, diagrams) with correct frontmatter, folder structure, and best practices. Use when user asks to "document a service", "document an agent", "document an AI agent", "create EventCatalog files", "add an event to the catalog", "document my architecture", "generate catalog documentation", "create documentation for my microservice", "document a database", "create an ADR", "document a data product", or "document an entity". |
| license | BUSL-1.1 |
| metadata | {"author":"eventcatalog","version":"1.0.0"} |
EventCatalog Documentation Skill
This skill tells and agent how to generate EventCatalog documentation files (services, agents, events, commands, queries, domains, flows, channels, containers, ADRs, data products, entities, diagrams) with correct frontmatter, folder structure, and best practices.
If the catalog directory already has resources, read the existing files to understand:
- Naming conventions (PascalCase IDs? kebab-case?)
- Folder structure (nested under domains or flat?)
- Which owners/teams are already defined
- Badge styles and patterns used
- Schema formats in use (JSON Schema, Avro, etc.)
Match new documentation to these existing conventions.
This ensures new documentation is consistent with what's already in the catalog.
Pull Request Review Workflow
When using this skill to update EventCatalog documentation from a source-code pull request, follow this procedure before writing files:
- Read the actual diff hunks, not only the current file contents. The diff shows what this pull request introduced, removed, or changed.
- Identify the pull request intent in domain terms: new event, changed event payload, new command, new query, changed service relationship, new endpoint, renamed message, deleted capability, or documentation-only change.
- Classify each change with a consistent type. When an existing event/command/query's payload or schema changes (for example a field is added to its published message), the change is
schema-changed — not event-added or service-changed. Use *-added only for a brand-new resource and *-removed when one is deleted.
- Map only those changed concepts to EventCatalog resources.
- Check whether each mapped resource already exists before creating a new one.
- Build a small edit plan that lists the exact catalog resource folders or files to change.
- Do not edit catalog files outside that plan.
Common PR-review mappings:
- A new event constant plus publish code means create or update that event resource and update the producing service
sends.
- A changed event payload means update that event schema, examples, and summary where needed.
- A renamed or fixed message ID means update the existing message relationship; do not create a new resource unless the new message is genuinely new.
- A new command/query handler or endpoint means create or update the matching command/query resource and service
receives/queries relationship where the catalog models it that way.
- A service that starts using a data store (a database, cache, or object store that holds state — e.g. new code that opens a DB connection pool, runs SQL, or reads/writes a cache) means create a container for that data store under the service (
services/{Service}/containers/{container}/index.mdx) with a container_type and technology, and map the service to it: add the data store id to the service's writesTo when it persists/updates data and readsFrom when it queries data (both when it does both). A connection string or client appearing in a changed file is enough to document the container only when this pull request introduced that data-store usage.
- A channel constant appearing in a changed file is not enough to create a channel. Only create or update channel resources when the channel itself was introduced, renamed, or its routing changed in the pull request.
- Message-transport infrastructure (a Kafka topic, message queue, event bus, SNS/SQS, Pub/Sub subject, etc.) is a CHANNEL, never a container/data store. When a service starts publishing or consuming messages over such transport, create a
channel (with the relevant protocols, e.g. kafka) and wire the service's sends[].to / receives[].from to it — do NOT model it as a container with writesTo/readsFrom. Containers are only for stateful stores (databases, caches, object stores) that hold data, not for moving messages.
- Existing catalog issues discovered during review but not introduced by the pull request are out of scope. Report them, but do not edit them.
- If no EventCatalog documentation changes are needed, return a clear no-op rationale.
Understanding format of resources
Generate files following the resource-specific references. Consult the appropriate reference file for the resource type:
references/services.md — Services with sends/receives, channel routing, containers
references/agents.md — Agents with model metadata, tools, sends/receives, containers, and flows
references/events.md — Events with schemas, payload examples, producer/consumer code
references/commands.md — Commands with REST operations and schemas
references/queries.md — Queries with REST operations and response schemas
references/domains.md — Domains with subdomains, services, and business context
references/flows.md — Business flows with steps, branching, and external systems
references/channels.md — Channels with routing, protocols, and parameters
references/containers.md — Containers (stateful data stores: databases, caches, object stores) with data classification
references/adrs.md — Architecture decision records with status, date, decision makers, appliesTo, and relationships
references/data-products.md — Data products with inputs, outputs, data contracts, lineage, and SLAs
references/entities.md — DDD/domain entities with identifiers, properties, relationships, and aggregate roots
references/diagrams.md — Reusable diagram resources (Mermaid, PlantUML, architecture diagrams)
references/ubiquitous-language.md — Ubiquitous language terms per domain (DDD glossary/dictionary)
references/teams-and-users.md — Teams and users (ownership)
references/components.md — Components (NodeGraph, Schema, Mermaid, Tabs, etc.) and resource references ([[type|Name]] wiki-style links)
references/supporting-collections.md — Changelogs, resource docs, custom docs, schemas, and Studio designs
Every resource file MUST include:
- Valid YAML frontmatter between
--- delimiters
id field matching existing catalog conventions
name as human-readable display name
version as semantic version string
summary as a concise 1-2 sentence description
CRITICAL: Always use index.mdx as the filename for versioned resources (services, agents, events, commands, queries, domains, flows, channels, containers, ADRs, data products, entities, diagrams). Teams and users use {id}.mdx files directly. Ubiquitous language uses ubiquitous-language.mdx. Place files in the correct folder path following the nested structure pattern:
domains/{DomainName}/services/{ServiceName}/events/{EventName}/index.mdx
domains/{DomainName}/agents/{AgentName}/index.mdx
domains/{DomainName}/data-products/{DataProductName}/index.mdx
domains/{DomainName}/entities/{EntityName}/index.mdx
domains/{DomainName}/diagrams/{DiagramName}/index.mdx
Or flat structure if the catalog uses that pattern:
services/{ServiceName}/index.mdx
agents/{AgentName}/index.mdx
events/{EventName}/index.mdx
adrs/{adr-id}/index.mdx
data-products/{DataProductName}/index.mdx
entities/{EntityName}/index.mdx
diagrams/{DiagramName}/index.mdx
Do not generate schemas collection entries directly. Generate or reference schema files from events, commands, or queries using schemaPath or schemas; EventCatalog creates the schemas collection from those references.
Validation
Before presenting the files to the user, verify:
- YAML frontmatter has
--- delimiters on both sides
- All
id fields are consistent (no spaces, match folder name)
- All
version fields are valid semver strings (e.g., 0.0.1)
- All message references in
sends/receives include id and optionally version
- Channel routing uses
to/from fields correctly in sends/receives
- Schema files referenced in
schemaPath actually exist or are generated
<NodeGraph /> component is included for architecture visualization
- Owner IDs reference real teams/users in the catalog
- Run the EventCatalog linter after making documentation changes and fix any errors before finishing
Common Patterns
Documenting a Service That Processes Messages
- Generate the service
index.mdx with receives and sends arrays
- If messages flow through channels, add
to/from fields to the sends/receives
- Generate each event
index.mdx if they don't already exist in the catalog
- Include
<NodeGraph /> in the service body to show message flow
- Generate related entities if the service owns important domain objects
- Add example payload sections for each message
- Place files in the correct nested folder structure
Documenting an Agent
When a user says "document my support agent that reads order data and uses Zendesk":
- Generate the agent
index.mdx with model, tools, receives/sends, readsFrom/writesTo, and flows where known
- Generate or reference events/commands/queries the agent consumes or produces
- Generate containers for data stores the agent reads or writes
- Include
<AgentTools /> when tools are documented
- Include
<NodeGraph /> so the agent appears in architecture visualizations
- If the agent belongs to a domain, add it to the domain's
agents frontmatter
Documenting a Domain
CRITICAL: A domain MUST have at least one service or agent. Never create an empty domain. If the user describes a domain, ensure services or agents are identified and generated for it.
When a user wants to document a full domain:
- Identify the services and agents that belong to this domain. If the user hasn't specified any, ask them: "What services or agents belong to this domain?" Do NOT create an empty domain.
- Generate the domain
index.mdx with the services field listing every service and the agents field listing every agent
- Include
entities, data-products, flows, and diagrams fields when those resources belong to the domain
- Generate each service and agent within the domain
- Generate each message referenced by the services and agents
- Generate entities, data products, diagrams, and channels if the user describes them
- Use the nested folder structure:
domains/{Domain}/services/{Service}/events/{Event}/, domains/{Domain}/agents/{Agent}/, domains/{Domain}/entities/{Entity}/, and domains/{Domain}/data-products/{DataProduct}/
- Generate a
ubiquitous-language.mdx file for the domain by extracting domain-specific terms from service names, agent names, event/command names, entities, and business processes. Place it at domains/{Domain}/ubiquitous-language.mdx. See references/ubiquitous-language.md for format and examples.
- CRITICAL: After generating all files, verify the domain's frontmatter
services field lists every service and agents lists every agent that belongs to it. Every service or agent created under a domain MUST be referenced in the domain's index.mdx:
services:
- id: OrdersService
- id: InventoryService
- id: PaymentService
agents:
- id: OrderSupportAgent
If a service or agent is nested inside the domain folder but not listed in the domain's frontmatter, it will not appear as part of that domain. Always cross-check.
Documenting an ADR
When a user describes an architecture decision:
- Generate
adrs/{adr-id}/index.mdx
- Use one of the supported statuses:
proposed, accepted, rejected, deprecated, or superseded
- Include a
date in YYYY-MM-DD format
- Add
decisionMakers and owners using existing team/user IDs where known
- Use
appliesTo to link the decision to impacted resources (service, event, domain, flow, data-product, entity, etc.)
- Use
supersedes, supersededBy, amends, amendedBy, or related when linking ADRs together
- Structure the body with
Context, Decision, and Consequences
Documenting a Data Product
When a user describes analytics, reporting, BI, ML feature, or derived operational data:
- Generate
data-products/{DataProductName}/index.mdx or nest it under the relevant domain/subdomain
- Add
inputs for upstream messages, services, containers, channels, or other resources
- Add
outputs for produced messages, services, containers, channels, or contracts
- If an output has a data contract, include
contract.path, contract.name, and contract.type
- Include
<NodeGraph /> and any relevant <SchemaViewer /> for contract files
- Document lineage, freshness, ownership, access patterns, and SLAs
Documenting an Entity
When a user describes a domain model, aggregate, data object, or business concept with properties:
- Generate
entities/{EntityName}/index.mdx, domains/{Domain}/entities/{EntityName}/index.mdx, or services/{Service}/entities/{EntityName}/index.mdx depending on catalog structure
- Include
identifier and aggregateRoot: true when applicable
- Add
properties with name, type, required, and description
- Use
references, referencesIdentifier, and relationType for relationships to other entities
- Include
<EntityPropertiesTable /> in the body to render the property table
- Link entities from domain/service frontmatter using
entities
Documenting a Diagram
When a user provides or asks for a reusable architecture, sequence, flow, or model diagram:
- Generate
diagrams/{DiagramName}/index.mdx or nest it under the relevant domain/subdomain
- Include
id, name, version, and summary
- Put the diagram in the body as a fenced
mermaid, plantuml, or other supported diagram block
- Reference the diagram from related resources using the
diagrams frontmatter field
Documenting a Business Flow
When a user describes a multi-step process:
- Identify distinct steps (user actions, service calls, message exchanges, external systems)
- Generate the flow
index.mdx with steps array
- Each step should have
id, title, and appropriate type (actor, service, agent, message, externalSystem)
- Connect steps with
next_step or next_steps for branching
Documenting Channel Routing
When a user describes how messages flow through infrastructure:
- Generate channel
index.mdx files with routes for channel-to-channel routing
- Update service or agent
sends/receives with to/from fields pointing to channels
- The full picture should show: Service or agent sends → Channel → routes to → Channel → service or agent receives
Quality Checklist
- Take your time to do this thoroughly
- Quality is more important than speed
- Do not skip validation steps
Before delivering documentation to the user, verify every file against this checklist:
- Frontmatter has valid YAML between
--- delimiters
id matches the folder name
version is a valid semver string
summary is concise and meaningful (not generic)
- Message relationships (
sends/receives) include id
- Channel routing (
to/from) references valid channel IDs
- Body includes
<NodeGraph /> for visualization when the resource has graph relationships
- Schema references point to real files
- Folder structure follows catalog conventions
- No duplicate resources (checked against existing catalog)
- Versioned resources use
index.mdx (or match the catalog's existing .md/.mdx convention); teams and users use {id}.mdx; changelogs use changelog.mdx/changelog.md
- Every domain has at least one service or agent — never create an empty domain
- Domain
services and agents frontmatter lists every service and agent that belongs to that domain
- Domain
entities, data-products, flows, and diagrams frontmatter lists nested resources when present
- Every domain has a
ubiquitous-language.mdx file with relevant domain terms extracted from services, agents, events, commands, entities, data products, and business processes
- ADRs have a valid status, date, decision makers when known, and
appliesTo references for impacted resources
- Data product contract files referenced in
outputs.contract.path exist when generated