Skip to content

Document agentic usage attribution in Audit Trail#37955

Draft
anthony-dgx wants to merge 3 commits into
masterfrom
anthony.dagneaux/document-agentic-audit-trail
Draft

Document agentic usage attribution in Audit Trail#37955
anthony-dgx wants to merge 3 commits into
masterfrom
anthony.dagneaux/document-agentic-audit-trail

Conversation

@anthony-dgx

Copy link
Copy Markdown
Contributor

Summary

  • Document the new @evt.actor.mode, @delegator.actor, and is_mcp_call fields on the Audit Trail events reference page, following the "Surface Agentic Usage In Audit Trail" GA announcement.
  • Add a "View a user's full activity, including agent-delegated actions" example to the Audit Trail overview page, covering the direct-actions-only vs. delegated-actions-included query distinction.
  • Add a new guide, Track Agentic Usage in Your Organization, covering: viewing all agent activity, viewing MCP-originated actions, tracing delegated user activity, and distinguishing agent vs. human events in the UI.

Notes

  • Coverage-gap and GovCloud caveats from the announcement are called out as Note callouts rather than presented as fully complete, since attribution is rolling out per agent team.
  • The per-product event tables on the events page are rendered from data/audit_events.json in websites-sources, not this repo — any brand-new discrete event types (vs. the cross-cutting actor/delegation fields added here) would need a separate change there.

Test plan

  • Vale linter passes on Files changed tab
  • Preview build renders new guide page and updated sections correctly
  • Links resolve (overview page, events page, new guide, anchors)
  • Docs team review for launch-readiness alongside the GA date

Add a new-field reference for @evt.actor.mode / @delegator.actor / is_mcp_call
on the events page, a delegated-activity query example on the Audit Trail
overview page, and a new guide on tracking agentic usage across the org,
following the GA feature announcement for agent actor + delegation attribution.

## View a user's full activity, including agent-delegated actions

When a Datadog AI agent acts on a user's behalf, the agent becomes `@evt.actor` for that event, and the user moves to `@delegator.actor`. Searching only `@usr.name:<USER_EMAIL>` returns the actions the user performed directly, and does not include actions a Datadog AI agent performed on their behalf.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We might want to be a little more specific and mention that the agent now fills in the values of @usr.*, while the original user fills in @delegator.*? This should be easier to filter on than only the @delegator.actor field.

To retrieve a user's complete activity, including delegated agent actions:

1. Navigate to [Audit Trail][1].
2. In the search bar, enter the query: `(@usr.name:<USER_EMAIL> OR @delegator.name:<USER_EMAIL>)`. Replace `<USER_EMAIL>` with the user's email address.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@delegator.name is the actual full name. If users want to filter on the email address, I would advise @delegator.id.

1. Navigate to [Audit Trail][1].
2. In the search bar, enter the query: `@evt.actor.mode:*`.

This query returns every event where a Datadog AI agent is the actor, whether the agent acted `interactive` (a user was present and approved the action) or `autonomous` (the agent acted on a system trigger, with no user present).

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
This query returns every event where a Datadog AI agent is the actor, whether the agent acted `interactive` (a user was present and approved the action) or `autonomous` (the agent acted on a system trigger, with no user present).
This query returns every event where a Datadog AI agent is the actor, whether the agent acted `interactive` (a user was present and approved the action) or `autonomous` (the agent acted on a system trigger, with no user involved).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants