Generating Custom Output

The AI agent can generate custom output files during a session — documents, reports, data tables, and more. These files are saved to the deliverables/ folder in the agent’s sandboxed filesystem and automatically uploaded as artifacts when the session ends.

How It Works

The agent always produces a session report summarizing what happened during the run. Additionally, you can request custom output either explicitly in your prompt or implicitly through task requirements.

Files in the deliverables/ folder are uploaded alongside other run artifacts (device logs, crash logs, etc.) and can be downloaded from the run details page.

Supported Formats

Examples

Explicit Requests

Ask the agent directly to create a specific deliverable:

Create a Design Document (pdf) covering the new clan feature.

Read survey.pdf and put your answers in answers.txt.

Implicit Data Collection

Have the agent track data as part of a broader task:

Play through all tutorial levels. For each level, track success/failure,
completion time, and number of attempts in level-stats.csv.

Test the checkout flow with 10 different product combinations.
Record each result in checkout-tests.csv with columns: products, total, success, notes.

Combined with Knowledge Files

Use knowledge files as input and generate output based on them:

Review the game against the attached design-spec.pdf.
Create a compliance checklist in compliance.md noting which requirements pass or fail.

Create 10 nexus test cases based on the acceptance criteria in user-story.md.

Best Practices

• Be specific about format — If you need a particular file type, mention it explicitly (e.g., “as a CSV” or “in PDF format”)
• Name your files — Suggest filenames in your prompt to keep outputs organized
• Define structure for data — When collecting metrics, specify the columns or fields you want tracked
• Keep deliverables focused — One file per purpose is easier to work with than one large file with everything