Introduction
The Web Function Package specification simplifies endpoint interaction by standardizing their description and access. This defined structure ensures all endpoints are consistent, discoverable, and self-documented.
Package definition
A package is a JSON object that contains the following keys:
| Key and type | Description |
|---|---|
base_url string | Required. Base URL for the package. MUST use the HTTP or HTTPS scheme. |
event_source_url string | Optional. An event source URL from which to receive server-sent-events. See the event specification for more information. |
pipeline_url string | Optional. A function pipelining URL. See the pipelining specification for more information. |
name string | Optional. The name of the package. |
flags array of strings | Optional. List of package flags. See the available flags section below for a complete list of flags available at the package level. |
version string | Optional. The version that this package describes. An opaque string. MUST be present when the versioned flag is set. See the versioning specification for more information. |
versions array of strings | Optional. The versions that are available. Each entry is an opaque string. MUST be present when the versioned flag is set. See the versioning specification for more information. |
docs string | Optional. Top-level documentation for the package. MUST be formatted as markdown. |
endpoints array of objects | Required. An array of endpoint definitions, as described below. |
events array of objects | Optional. An array of events definitions, as described below. See the event specification for more information. |
errors array of objects | Optional. A list of common ERROR_CODEs that can be returned by any endpoints in this package. Note that clients SHOULD only refer to this list if an endpoint uses the error_triple flag. See the error specification for more information. |
objects array of objects | Optional. A list of named object definitions, as described below. Each definition can be referenced as a refined object type (object.<name>) anywhere a type is expected. See the types section for more information. |
Types
Everywhere a type is expected—an endpoint's returns, or an argument's or attribute's type—it is expressed using the following grammar.
A type is one of:
- A base JSON type, written as a bare string:
object,array,string,number,boolean, ornull. - The special
anytype, which represents any base JSON type.anyMUST NOT be refined. - A refined type, written as a base type and a refinement joined by a dot, for example
number.u32,string.email, orobject.user. See the refinements section for the complete list. Anobject.<name>refinement references a definition in the package'sobjectsarray. - An array type, written as a JSON array whose entries are the types that the array's items may take (a union). This is fully recursive: an array's items may themselves be array types. An array type MUST NOT be empty; the bare string
arrayand["any"]are equivalent and denote an array ofany.
A returns or type value is either a single type or a union of types written as a non-empty JSON array. When it is an array, the outermost array is the union itself, so an entry that is an array is an array type. Empty arrays are not permitted anywhere: use the bare any (equivalently ["any"]) to match a value of any type, and the bare array (equivalently ["array"]) to match an array of any. A nested array appearing as an entry is an array type, as described above.
For example, ["string", ["object.user"]] represents a value that is either a string or an array of object.user, while the bare "object.user" is a single user object and "any" matches any value.
Endpoint definition
Each element in the endpoints array is an object that contains the following keys:
| Key and type | Description |
|---|---|
name string | Required. Suffix for the endpoint URL, appended to the PACKAGE_BASE_URL to form the full endpoint URL. Endpoint names MUST be unique within a package; overloading (two endpoints sharing the same name) is not permitted. |
returns string or array | Required. The type(s) returned by the endpoint, expressed using the type grammar: either a single type (for example "object.user") or a non-empty array representing a union. Entries may be base types, any, refined types (for example number.u32 or object.user), or nested arrays representing array types. |
flags array of strings | Optional. List of endpoint flags. See the available flags section below for a complete list of flags available at the endpoint level. |
group string | Optional. A name used to categorize or group similar endpoints together. This should be used by documentation tools to organize related endpoints. |
docs string | Optional. Documentation for the endpoint. MUST be formatted as markdown. |
errors array of objects | Optional. A list of specific ERROR_CODE that can be returned by this endpoint. Note that clients SHOULD only refer to this list if an endpoint uses the error_triple flag. See error for more information. |
arguments array of objects | Required. Arguments required by the endpoint, as described below. If no arguments are required, the array should be empty. |
attributes array of objects | Optional. Attributes of the object returned by the endpoint, as described below. This element MAY be present only when returns is (or includes) the bare object type. |
URL composition
The full URL of an endpoint is formed by joining the package's base_url and the endpoint's name with exactly one / separator:
- If
base_urlends with a/, thenameis appended directly. - If
base_urldoes not end with a/, a single/is inserted before thename. - An endpoint
nameMUST NOT begin or end with a/.
For example, given a base_url of https://api.example.com (or, equivalently, https://api.example.com/) and an endpoint name of find-user-by, the full endpoint URL is https://api.example.com/find-user-by.
Event definition
Each element in the events array is an object that contains the following keys:
| Key and type | Description |
|---|---|
name string | Required. The name of the event. |
group string | Optional. A name used to categorize or group similar events together. This should be used by documentation tools to organize related events. |
docs string | Optional. Documentation for the event. MUST be formatted as markdown. |
attributes array of objects | Required. Attributes in the event object, as described below. |
Argument definition
Each element in the arguments array is an object that contains the following keys:
| Key and type | Description |
|---|---|
name string | Required. Name of the argument. |
type string or array | Required. The type(s) accepted for the argument, expressed using the type grammar: either a single type (for example "string.email") or a non-empty array representing a union. Entries may be base types, any, refined types (for example number.u32 or object.user), or nested arrays representing array types. |
group string | Optional. A name used to categorize or group similar arguments together. This should be used by documentation tools to organize related arguments. |
choices array | Optional. An array specifying the exact, case-sensitive values that are permitted for this argument. Each value in the choices array must conform to one of the types specified in the argument's type key.Note that if the argument's type is (or includes) the array type, choices may contain strings or numbers representing the allowed values that can be included in the array. |
flags array of strings | Optional. List of argument flags. See the available flags section below for a complete list of flags available at the argument level. |
docs string | Optional. Description of the argument. MUST be formatted as markdown. |
Attribute definition
Each element in the attribute array is an object that contains the following keys:
| Key and type | Description |
|---|---|
name string | Required. Name of the attribute. |
type string or array | Required. The type(s) of the attribute, expressed using the type grammar: either a single type (for example "string.email") or a non-empty array representing a union. Entries may be base types, any, refined types (for example number.u32 or object.user), or nested arrays representing array types. |
values array | Optional. An array specifying the exact, case-sensitive values that could be present in this attribute. Each value in the values array must conform to one of the types specified in the attribute's type key.Note that if the attribute's type is (or includes) the array type, values may contain strings or numbers representing the allowed values that could be part of the array. |
flags array of strings | Optional. List of attribute flags. See the available flags section below for a complete list of flags available at the attribute level. |
docs string | Optional. Description of the attribute. MUST be formatted as markdown. |
Object definition
Each element in the objects array is an object that contains the following keys:
| Key and type | Description |
|---|---|
name string | Required. Name of the object. It is referenced as a refined object type (object.<name>). Object names MUST be unique within a package. |
arguments array of objects | Optional. The object's members when it is referenced in an argument context, each following the argument schema. MUST be present when the object is referenced in an argument context. |
attributes array of objects | Optional. The object's members when it is referenced in an attribute context, each following the attribute schema. MUST be present when the object is referenced in an attribute context. |
An object.<name> reference appears in one of two contexts, which determines which set of definitions applies:
- Argument context — the object is referenced as an argument's
type. The object'sargumentsdescribe its members. - Attribute context — the object is referenced as an endpoint's
returnsor as an attribute'stype. The object'sattributesdescribe its members.
The context propagates recursively: when an argument's or attribute's type references another object, that nested object is interpreted in the same context as the definition that references it. Because an object MAY be referenced in both contexts within the same package, it MAY define both arguments and attributes; each set is used only in its matching context.
Error definition
Each element in the error array is an object that contains the following keys:
| Key and type | Description |
|---|---|
code string | Required. Machine readable code for the error. |
docs string | Optional. Description of the error. The description SHOULD focus on explaining all the situations in which this error will arise. MUST be formatted as markdown. |
Available flags
The Type column indicates the level at which each flag may appear. A flag MUST only be used at its designated level (package, endpoint, argument, or attribute) and MUST NOT appear at any other level.
| Name | Type | Description |
|---|---|---|
versioned | Package | Indicates the package is versioned using the Api-Version header. See the versioning specification for more information. |
package | Endpoint | Indicates the endpoint returns another package definition. |
event_source | Endpoint | Indicates the endpoint returns an event source URL. An endpoint with this flag MUST declare a returns of ["string"]. See the event specification for more information. |
error_triple | Endpoint | Indicates the endpoint adheres to the error response structure as per the error specification. When this flag is set and the response status code is 400, the response body MUST be an array (the error triple), regardless of the endpoint's declared returns type. See the error specification for more information. |
bearer_auth | Endpoint | Indicates the endpoint requires authentication via a bearer token. |
capture_bearer | Endpoint | Indicates the endpoint will return a bearer token in its response. See the authentication specification for more information. |
paginated | Endpoint | Indicates the endpoint supports pagination. |
private | Endpoint | Indicates the endpoint is intended for internal use and is not part of the public API. Documentation tooling SHOULD omit endpoints with this flag from generated or published documentation. |
required | Argument | Indicates if the argument is mandatory. |
nullable | Attribute | Indicates the attribute is optional: its key MAY be absent from the object, and when the key is present its value MAY be null. Consumers SHOULD treat a missing key and a null value equivalently. |
Refinements
A refinement narrows a base JSON type, providing more specific constraints than the base JSON types alone. Refinements are particularly useful for languages with strict typing. A refined type is written by joining the base type and the refinement with a dot, for example number.u32 or string.email.
The any type MUST NOT be refined. In addition to the refinements listed below, the object type may be refined with the name of a definition in the package's objects array (object.<name>).
| Refinement | Base JSON Type | Explanation |
|---|---|---|
u32 | number | Unsigned 32-bit integer. |
u64 | number | Unsigned 64-bit integer. |
i32 | number | Signed 32-bit integer. |
i64 | number | Signed 64-bit integer. |
f32 | number | 32-bit floating point number. |
f64 | number | 64-bit floating point number. |
timestamp | number | A UNIX timestamp (seconds since the epoch). |
date | string | A date string in ISO 8601 format (e.g., YYYY-MM-DD). |
time | string | A time string in ISO 8601 format (e.g., HH:MM:SS). |
datetime | string | A date-time string in ISO 8601 format. |
uuid | string | A Universally Unique Identifier. |
base64 | string | A Base64 encoded string. |
email | string | An email address. |
phone | string | A phone number in E.164 format. |
url | string | A Uniform Resource Locator. |
uri | string | A Uniform Resource Identifier. |
ipv4 | string | An IPv4 address. |
ipv6 | string | An IPv6 address. |
hostname | string | A valid hostname as defined by RFC 1123. |
Validation requirements
The constraints expressed in the definition tables above—required versus optional keys, value types, allowed values, and flag scoping—are themselves normative and MUST be enforced.
In addition, each package MUST pass the following validations:
- The base URL MUST be valid as per
RFC 3986. - All specified types and flags MUST match defined constraints.
- Every type in a
returnsortypevalue MUST be a valid base type,any, or a valid refinement. Arrays MUST NOT be empty at any depth; a union and every array type MUST contain at least one entry. - The
anytype MUST NOT be refined. - Each
object.<name>refinement MUST resolve to an entry in the package'sobjectsarray. - An object referenced in an argument context MUST define an
argumentsarray; an object referenced in an attribute context MUST define anattributesarray. - Object
names MUST be unique within the package.
Security considerations
Several keys in a package—the docs field on the package, endpoints, events, arguments, attributes, and errors—contain Markdown authored by the API developer. Because a package may be fetched from a third party or generated dynamically, tooling that renders these docs fields MUST treat them as untrusted input. Renderers MUST neutralize embedded scripting and active content (for example, by sanitizing the generated HTML) to avoid cross-site scripting (XSS) and similar injection attacks.
Example
A simple example of a Web Function Package defining a single endpoint might look like this:
{
"base_url": "https://api.example.com",
"name": "ExamplePackage",
"flags": [],
"docs": "This package defines endpoints for Example API.",
"errors": [],
"objects": [
{
"name": "user",
"attributes": [
{
"name": "id",
"type": "string.uuid"
},
{
"name": "email",
"type": "string.email"
}
]
}
],
"endpoints": [
{
"name": "find-user-by",
"returns": "object.user",
"flags": [],
"group": "users",
"docs": "Retrieves user data.",
"errors": [],
"arguments": [
{
"name": "id",
"type": "string.uuid",
"choices": [],
"flags": ["required"],
"docs": "Identifier of the user."
}
]
}
]
}
Retrieval
A package can reach a client in more than one way, but only one of them is governed by this specification.
Invocation response (covered). A Web Function Package MAY be returned as the response body of a Web Function Endpoint invocation, that is, an HTTP
POSTas described in the endpoint specification. This is the only retrieval method defined by this specification. It enables a dynamic and recursive discovery mechanism where endpoints return packages as part of their operation, supporting complex scenarios where packages are interdependent or where the structure of available APIs can change.Side-loaded (out of scope). A package MAY also be obtained out of band, for example distributed as a static file read from the filesystem, or fetched with a plain HTTP
GET. Such a side-loaded package MAY be static or dynamically generated by a web application so its contents adapt to the request or to application state. These mechanisms are common and perfectly valid, but they are not part of this specification: how a side-loaded package is located, transported, or cached is left entirely to the implementation.