diff --git a/content/departments/engineering/teams/core-services/sams/index.md b/content/departments/engineering/teams/core-services/sams/index.md
index 604c2d30216e..69c120e44fe4 100644
--- a/content/departments/engineering/teams/core-services/sams/index.md
+++ b/content/departments/engineering/teams/core-services/sams/index.md
@@ -13,6 +13,10 @@ It is compliant with [OAuth 2](https://oauth.net/2/) and [OIDC](https://openid.n
 
 The [OpenID Discovery](https://accounts.sourcegraph.com/.well-known/openid-configuration) endpoint lays out all the protocol details that a Service Provider (aka. Relay Party) needs to know to integrate with SAMS.
 
+## System designs
+
+- [Token scope specification](./token_scope_specification.md)
+
 ## Security measures
 
 Here is a list of security measures that are notable to systems integrating with SAMS:
diff --git a/content/departments/engineering/teams/core-services/sams/token_scope_specification.md b/content/departments/engineering/teams/core-services/sams/token_scope_specification.md
new file mode 100644
index 000000000000..3d15d35a2ca9
--- /dev/null
+++ b/content/departments/engineering/teams/core-services/sams/token_scope_specification.md
@@ -0,0 +1,37 @@
+# SAMS Token Scope Specification
+
+This document formally defines:
+
+1. Components of a token scope, i.e. how do token scopes look?
+1. Token scope matching strategy, i.e. how authorization works?
+
+## Basic format
+
+A scope is always consists of three parts: **service**, **permission hierarchy**, and **action**, and are separated by double colons `::`, e.g. `sams::user.roles::read`, `ssc::subscriptions::read`:
+
+1. The total length of a scope cannot exceed 255 characters.
+1. **Service**: the slug of the service name, up to 30 characters.
+    - It can only contain `[a-z_]` characters, e.g. `sams`, `ssc`, `dotcom`, `cody_gateway`.
+1. **Permission hierarchy**: the hierarchy of the permissions under the given service, up to 215 characters.
+    - It can only contain `[a-z_\.]` characters, e.g. `user`, `user.metadata`, `user.roles`.
+    - There is no conceptual limit for the number of hierarchy levels (other than the scope length limit).
+    - Every hierarchy grants access to all sub-permissions (aka. prefix matching), e.g.:
+        - `user` grants `user.roles`, `user.metadata`, etc.
+        - `user.metadata` grants `user.metadata.cody`, `user.metadata.dotcom`, etc.
+        - If an access token is granted with `<service>::user.profile::write`, the client is able to write to ALL aspects of the user profile. If a service wants to only allow access to a specific section, the service will need to create more specialized scopes. e.g. granting both `<service>:user.profile.avatar_url::write` and `<service>::user.profile.display_name::write`.
+    - The hierarchy design also helps us to avoid unnecessarily updating all services when a new sub-permission is introduced.
+    - It is by design that “access-to-everything” of a service is explicitly disallowed, therefore no notion provided for it.
+1. **Action**: the allowed action for the given permission of the given service, up to 10 characters.
+    - It is a predefined/hardcoded list of values, it has no point being dynamic (strings) for being conservative:
+        - `read`: allowed to read resources from the service
+        - `write`: allowed to create or make updates to the resources of the service
+        - `delete`: allowed to delete resources on the service
+    - Every `(permission, action)` must be granted explicitly, there is no action hierarchy, e.g. having write permission doesn’t automatically grant read permission.
+
+## Scope alias
+
+This is initially intended for making OIDC scopes be compatible with our design (`profile` -> `sams::user.profile::read`), but could also be a useful extension in the future, e.g. in the event of scope renaming efforts.
+
+## References
+
+- [SAMS token scope specification draft](https://docs.google.com/document/d/1ixjjEE9AhSt-ea_1J2yoRENwCEVZArRRqFxUVkjjYW8/edit)