desktop
I Instagram

Model

Tasks

Reply to a product DM Update profile link Collect creator candidates Respond to campaign comments Prepare a post draft
1 2 3 4 5
RL env

Instagram environment

5 tasks with a model prompt, seeded environment state, and grader contract.

Task 1

Reply to a product DM

Prompt
Open the unread DM asking about product IG-882 and reply with the size availability from the brief.
Environment
The inbox includes several unread DMs and one product availability question.
Grader
Checks the correct conversation reply and product size value.
Task 2

Update profile link

Prompt
Replace the profile website link with the campaign URL from the task brief.
Environment
The profile edit page contains an outdated website URL.
Grader
Checks that the website field equals the campaign URL and other profile fields remain unchanged.
Task 3

Collect creator candidates

Prompt
Search the seeded hashtag and save three creator profiles that match the follower range in the brief.
Environment
Search results contain creators and brand accounts with visible follower counts.
Grader
Checks saved profiles count, follower range, and hashtag source.
Task 4

Respond to campaign comments

Prompt
Open the campaign post and reply to the two unanswered sizing questions with the approved response.
Environment
The campaign post has a mix of answered and unanswered comments.
Grader
Checks replies on the two target comments and no duplicate replies.
Task 5

Prepare a post draft

Prompt
Create a post draft using the selected asset, caption, and first comment from the brief.
Environment
The media library includes the selected asset and the composer has no current draft.
Grader
Checks asset selection, caption text, first comment, and draft status.
UseDesktop Evals

Computer-use agent evals.

RL envs Main site Docs Blog