Skip to main content
A skill is a packaged capability that an agent can use. You upload the package as a ZIP archive, MemoryLake runs a security review on it, and once the review passes you attach the skill to an agent by listing it in the agent’s skills. Skills belong to your tenant, not to a workspace: one skill can be attached to any agent in any of your workspaces.

​
How Skills Work

1

Request an upload slot

Call Request Package Upload. It returns a presigned upload_url, the upload_headers to send with it, and an s3_uri that names the upload.
2

Upload the ZIP archive

Send the archive as the body of a single PUT to upload_url, with every header from upload_headers.
3

Create the skill

Call Create Skill with the s3_uri in package_ref. The skill is created with its first version, and latest_security_status is pending.
4

Wait for the security review

Poll Get Skill until latest_security_status is no longer pending.
5

Attach it to an agent

Pass the skill’s id in skills when you create an agent or create an agent version.

​
Security Review

Every package is reviewed before an agent can use it. The result is reported per version, and the skill’s latest_security_status reports it for the latest version: latest_security_status can also be absent. Read latest_version to tell what that means:
  • latest_version above 0: security review is switched off on this deployment, and the skill can be referenced.
  • latest_version is 0: no package has been published, and the skill cannot be referenced.
  • latest_version absent too: the state could not be read, and any reference to the skill is refused until it can.

​
Referencing a Skill From an Agent

An agent references a skill with skill_id and, optionally, skill_version. Without skill_version the reference follows the latest version; with it, the reference is pinned to that version and checked against that version’s own review result. When you create an agent or an agent version, every skill it references is checked. If any of them cannot be used, the whole request fails with 400 and error code SKILL_NOT_USABLE, and the message names each skill that failed and why. Built-in skills provided by MemoryLake are checked the same way.
Publishing a new version of a skill puts it back under review. Until the new version passes, agents whose reference follows the latest version cannot be created or given new versions with that skill. Agent versions pinned to an older version number are unaffected.

​
Names and Identifiers

name must be unique within your tenant, and cannot be changed after the skill is created. id (for example skill-2b0f5f9f4ea54767822f51f4940c8f24) is what every other endpoint and every agent reference uses.

​
Built-in Skills

MemoryLake also provides built-in skills. Agents can reference them by id, but they are not returned by List Skills and cannot be updated or deleted.

​
Endpoints