Designing metadata event schema

Welcome to the new home of Slack developer docs!

We're still building and not all features are available quite yet. Enjoy this peek into the future!

Not ready for the future? Return to the past at api.slack.com.

This document provides design guidelines for developers looking to emit typed metadata events from their applications. These guidelines are intended to be strong recommendations (rather than prescriptive). Following these patterns will serve developers in providing better readability, maintainability, and DRYness across their workflows and applications.

While in the future this guide may serve as a launching point for additional features, none of these guidelines are enforced programmatically and only provide guidance in how to best use structured data.

This document repeatedly uses the word should (and occasionally MUST), as to be interpreted as described in [RFC2119].

General Guidelines

Metadata Events are structured data payloads that contain information about events occurring in your Slack-connected application, in the form of a custom event_payload as part of a message's metadata property.

Developers can use different Slack API methods to transmit these events. One such method would be chat.postMessage, where metadata can be included as a JSON payload, and arrives in channel as part of a message (not visible to the user).

Payload Structure

Metadata always includes an event_type and an event_payload. While order doesn't matter, developers should send event_type first for readability.
Event properties are always sent as part of an event_payload, and property order within the event payload is not important.
The value of event_type should be an alphanumeric string, and human readable. The value of this field may appear in the UI to developers, so keep this in mind when choosing a value. Developers should make an effort to name with the pattern <resource_name_singular> and <action_in_past_tense>

{
    "event_type": "job_created",
    "event_payload": {
        "job_id": 12345,
        "date_created": 1623346792
    }
}

A sample metadata payload as part of chat.postMessage.

Property Basics

When writing your JSON representation of event metadata:

Always use double quotes
- Property names should be surrounded with double quotes.
- Property values, if string, should be surrounded with double quotes, otherwise, no quotes should be used.
Names should always begin with a character [a-z], and may only contain characters, numbers [0-9] and underscores [_]
Property names should be an alphanumeric string and human readable. The value of this field may appear in the UI to developers, so keep this in mind when choosing a value.

    {
      "event_type": "example_created",
      'event_payload': { ❌ // Always use double quotes
        "user": 'U1234567', ❌ // Always use double quotes
        "date_created": "1623346792" ❌ // Number (epoch) should not have quotes
        "date_deleted": 1623346793 ✅
        "9tails": "gotta catch em all!", ❌ // Don't begin properties with a number
        "bad idea": true, ❌ // no spaces in property names
        "the_quick_brown_fox_jumped_over_the_lazy_dog": false, ❌ // too long
        "name": "Bologne" ✅
    }

Casing and Compounding

Property names should always be snake case (lowercase with underscores)

✅ Recommended	❌ Not Recommended
`user`	`User`
`user_id`	`userId`
`assigned_user_id`	`assignedUserId`

We should avoid using articles such as 'a', 'the', 'of' unless needed to convey meaning.
- e.g. names such as a_user, the_team should NOT be used, rather user, team should be preferred.
Multi-word properties should be separated by an underscore.

✅ Recommended	❌ Not Recommended
`poll_id`	`pollId`
`channel`	`the_channel`, `theChannel`, `Channel`

    "event_payload": {
      "user": "U1234567", ✅
      "date_created": 1623346792 ✅
      "poll_id": 42 ✅
    }

    "event_payload": {
      "User": "U1234567", ❌
      "DateCreate": 1623346792 ❌
      "the_poll_id": 42 ❌

Singular vs Plural Names

Plural names should always represent a type array. All other property names should be singular. (Including counts).
Properties should not contain superfluous words to indicate units of the item exist (i.e. user_objects (just users), or task_items (just tasks)

Context	✅ Recommended	❌ Not Recommended
Property representing an array of User strings.	`users`	`user`
Property representing an array of tasks	`tasks`	`task_items`
Property representing an object with a number of properties.	`profile`	`profiles`

✅ Recommended

    "event_payload":{
       "job":{
          "id":1001 ✅
       },
       "tasks":[
          {
             "id":420, ✅
             "name":"Task"
          },
          {
             "id":421,
             "name":"Task 2"
          }
       ],

❌ Not Recommended

    "event_payload": {
      "jobs":{ ❌
        "id": 1001,
      },
      "task_items":[{ ❌
      "id": 420,
      "name":"Task"
      },
      {     
      "id": 421,
      "name":"Task 2"   
    }],

Flattened data vs Structured Hierarchy

Data should only be grouped as it is sensible semantically. Developers should not arbitrarily group data for visual convenience. We currently don't support objects within objects (unless a custom type has already been defined), or arrays that contain objects or other arrays.

Properties should not use superfluous hierarchy to represent an ID unless there is a semantic use for the resulting object.

❌ Not Recommended

    "event_payload": {
      "user":{ ✅
        "id": 123
        "date_created": 1623857773,
        "names": { ❌ /* There is no reason to group these unless the object has a use */
          "first_name": "John",
          "middle_name": "Dino",
          "last_name": "Hammond",
        },
      }
    }

✅ Recommended

    "event_payload": {
      "user":{ ✅
        "id": 123,
        "date_created": 1623857773,
        "first_name": "John",
        "middle_name": "Dino",
        "last_name": "Hammond"
      }
    }

Flattened, an event containing metadata about a poll and some associated work called job might look like so:

    "event_payload": {
        "job_id": 12345,
        "date_created": 1623346792
        "poll_name": "my_poll",
        "poll_id": 42
        "poll_vote_count": 19
        "permalink_url": "https://example.com/my_poll/42"
    }

Using a structured hierarchy, a potentially more useful format could look like:

    "event_payload": {
        "job_id": 12345,
        "poll": {
          "date_created": 1623346792
          "name": "my_poll",
          "id": 42
          "vote_count": 19
          "permalink_url": "https://example.com/my_poll/42"
        }
    }

Using structured hierarchy where meaningful provides a number of benefits:

Reusability of objects across multiple applications and workflows
Clear intentionality of fields without overly verbose naming (i.e date_created).
Creating new object types

Null Values

Fields that are required should always have a value. Wherever possible, avoid sending null. Optional fields that are not sent as part of an event should be assumed to be null.

Consider configuring your event consumption in such a way that the setting and unsetting of specific properties is controlled by their presence in the event payload.

For example, if we wanted to set the first and last name, our event would look like so:

✅ Recommended

    "event_payload": {
      "first_name": "John",
      "last_name": "Hammond"
    }

If the user already existed, but had been previously set up with a middle name, and we wanted to unset the middle name, our event would look like so (the same):

✅ Recommended

    "event_payload": {
      "first_name": "John",
      "last_name": "Hammond"
    }

Anti-patterns:

❌ Not Recommended

    "event_payload": {
      "first_name": null, ❌ // never set required values to be null
      "last_name": "Hammond"
    }


    "event_payload": {
      "first_name": "John",
      "middle_name": null, ❌ // not recommended way to unset optional value
      "last_name": "Hammond"
    }

Type Selection

While most basic values can be defined using primitives, consider using custom types and objects wherever it may improve clarity, extensibility, and reusability.

Examples:

Representing a non Slack User ID (perhaps one associated with your customers?), declare your own type definition called mybusiness#/reference/objects/user-object_id as opposed to using a string or integer.
Creating a custom mybusiness#/reference/objects/user-object object will let you add new properties to the type object later without changing the event schema
Slack has many predefined types for you (non exhaustive). You should always use them over defining them yourself:
- slack#/reference/objects/channel-object
- slack#/reference/objects/channel-object_id
- slack#/reference/objects/user-object
- slack#/reference/objects/user-object_id
- slack#/types/team
- slack#/types/team_id
- slack#/types/timestamp

Property and Value Guidelines

DateTime

Properties

Date and time property keys should begin with date_.
DateTime properties should always use past tense in the format date_{past_tense_verb} like date_created, date_archived. The exception to this rule is when the value of the date going into the property is intended to generally be greater than the timestamp of the event.

✅ Recommended	❌ Not Recommended
`date_expired`	`date_expire`
`date_created`	`date_create`, `created_at`

If value that will go into the event is always greater than event timestamp at time of sending, than it should use future tense by dropping date and appending at

✅ Recommended	❌ Not Recommended
`expires_at`	`date_expires`
`send_at`	`date_send`
`ticket_due_at`	`date_ticket_due`, `ticket_due`

Properties describing the modification of a resource should include the resource noun in the name of the property when the property is not nested in an object that provides context. i.e. date_{resource}_{past_tense_verb}

✅ Recommended	❌ Not Recommended
`date_user_created`	`user_creation_date`
`date_task_deleted`	`task_date_deleted`

✅ Recommended

    "event_payload": {
      "date_project_created":1515472552, ✅
      "task":{
        "id": 1002,
        "date_created":1525472552, ✅
        "due_at": 1725472552 ✅
      },

❌ Not Recommended

    "event_payload": {
      "date_project_created":1515472552, ✅
      "task":{
        "id": 1002,
        "date_task_created": 1525472552, ❌
      },

Values

Date & time objects should be returned as integer unix epoch timestamp
Unix epoch timestamp values should be have second precision, not millisecond

✅ Recommended	❌ Not Recommended
`1525472552`	`Fri, 04 May 2018 22:22:31 GMT`
`1514764800`	`2018-01-01T00:00:00Z`

    types:
      date_created:
        description: Date object was created
        title: Date Created
        type: slack#/types/timestamps
        example: 1525472552

DateTime Relative

Properties

Properties denoting relationships in time should use following verbs:
- since, until - inclusive
- before, after - exclusive

✅ Recommended	❌ Not Recommended
`since_completed`	`since_complete`
`until_completed`	`until_complete`
`before_expired`	`time_to_expire`
`after_expired`	`time_after_expired`

Values

Relationships should be expressed in seconds as an integer

    since_deleted:
      description: How long its been since the issue was deleted
      title: Time Since Deleted
      type: integer
      example: 3600

DateTime Interval

Properties

When denoting intervals in time (such as calendar events), those periods should be described using two properties: date_<noun>_start **and date_<noun>_end.
Avoid the anti pattern of a combination of properties like date_start and then since_started to describe objects which have a start and an end (interval).

Values

Interval start and stop should be returned as integer unix epoch timestamp
Unix epoch timestamp values should be have second precision, not millisecond

Counts

Properties

Counts should be named as <singular_resource>_count
Counts represent a number, not a collection

✅ Recommended	❌ Not Recommended
`response_count`	`num_responses`
`ticket_count`	`num_tickets`
`authed_member_count`	`members_authed`
`item_count`	`total_items`
`user_count`	`users`

✅ Recommended

    "event_payload": {
                 "user_count": 10,

❌ Not Recommended

    "event_payload": {
                 "users": 10,

Values

Counts should always be represented by an integer.

    issue_count:
      description: How many issues are there
      title: Issue Count
      type: integer
      example: 12

Booleans

Properties

Booleans should be prefixed with is_ or has_

✅ Recommended	❌ Not Recommended
`is_expired`	`expired`
`is_due`	`due`
`has_attachments`	`attachments`

Values

Boolean values may only be true or false.

Enums

Properties

Enums should always be non plural

Context	✅ Recommended	❌ Not Recommended
Property representing an incident level	`severity`	`severities`

Values

Enum values should always be strings in uppercase

    severity:
      description: Severity of the incoming incident
      title: Incident Severity
      type: enum:["SEV1", "SEV2", "SEV3"],
      example: "SEV2"

    "metadata":{
      "event_type": "new_incident",
      "event_payload": {
                   "severity": "SEV2"

Identities

Properties that contain an ID should denote it with the suffix _id

    "event_payload": {
      "user_id": "U1234M112", ✅
    }

    "event_payload": {
      "user": "U1234M112" ❌
    }

Slack Types

Properties

Properties representing Slack objects (channel, user, team, etc) should be defined with Slack provided types slack#/types

✅ Recommended

    types:
      assignee_user_id:
        description: Slack user ID of assignee user
        title: Assignee user's ID
        type: slack#/reference/objects/user-object_id ✅
        example: U01234XM1

❌ Not Recommended

    types:
      assignee_user_id:
        description: Slack user ID of assignee user
        title: Assignee user's ID
        type: string ❌ // Use the slack user_id type
        example: U01234XM1

Slack Message Timestamps

Properties

Developers should call a property that represents a Slack message timestamp message_timestamp

✅ Recommended	❌ Not Recommended
`message_time stamp`	`timestamp`, `ts`

Values

Slack message timestamps MUST be sent as a string

✅ Recommended	❌ Not Recommended
`"1525471335.000320"`	`1525471335000320`
`"1525471335.000320"`	`1525471335`

✅ Recommended

    "metadata":{
      "event_type": "slack_message_processed",
      "event_payload": {
                   "date_received": 1525472552, ✅
                   "message_timestamp": "1525471335.000320" ✅

❌ Not Recommended

    "metadata":{
      "event_type": "slack_message_processed",
      "event_payload": {
                   "date_received": "2018-01-01T00:00:00Z", ❌
                   "message_timestamp": 1525471335 ❌

Latitude/Longitude

Latitude/Longitude should be strings formatted as recommended by ISO 6709.
They should favor the ±DD.DDDD±DDD.DDDD degrees format.

{
    "slack_hq": "+40.6894-074.0447"
}

Patterns

Here are some example patterns of types and events as they would be described by an app manifest file, and the resulting output of your events:

On-call Escalation

This is what an example message schema might look like if you were describing an event sent by a product like PagerDuty or similar.

Types - Manifest Schema

    types:
      incident_severity:
        type: string
        enum: ["S1", "S2", "S3", "S4"]
        choices:
          S1:
            title: Severity 1
          S2:
            title: Severity 2
          S3:
            title: Severity 3
          S4:
            title: Severity 4
      incident:
          title: Incident
          description: Represents an incident that was opened and hopefully resolved
          type: object
          required:
            - id
            - title
            - summary
            - severity
            - reported_by
            - created_at
          properties:
            id:
              type: integer
              title: Incident ID
            title:
              type: string
            summary:
              type: string
            severity:
              type: "#/types/incident_severity"
            reported_by:
              type: slack#/reference/objects/user-object_id
            assigned_to:
              type: slack#/reference/objects/user-object_id
              description: Who was on call and assigned to the incident
              nullable: true
            date_created:
              type: slack#/types/timestamp
            date_closed:
              type: slack#/types/timestamp
              nullable: true

Event - Manifest Schema

    events:
      incident_created:
        title: Incident Created
        description: Event triggered when a new incident is created
        type: object
        required:
          - incident
        properties:
          incident:
            type: "#/types/incident"

Metadata - Message Object

"metadata": {
    "event_type": "incident_created",
    "event_payload": {
        "incident": {
            "id": 9001,
            "title": "A new incident was created!",
            "summary": "Something really bad!",
            "severity": "S1",
            "reported_by": "U12345678",
            "date_created": 1623792755
        }
    }

General Guidelines​

Payload Structure​

Property Basics​

Casing and Compounding​

Singular vs Plural Names​

Flattened data vs Structured Hierarchy​

Null Values​

Type Selection​

Property and Value Guidelines​

DateTime​

DateTime Relative​

DateTime Interval​

Counts​

Booleans​

Enums​

Identities​

Slack Types​

Slack Message Timestamps​

Latitude/Longitude​

Patterns​

On-call Escalation​

General Guidelines

Payload Structure

Property Basics

Casing and Compounding

Singular vs Plural Names

Flattened data vs Structured Hierarchy

Null Values

Type Selection

Property and Value Guidelines

DateTime

DateTime Relative

DateTime Interval

Counts

Booleans

Enums

Identities

Slack Types

Slack Message Timestamps

Latitude/Longitude

Patterns

On-call Escalation