Scenarios
A scenario is a collection of prompts grouped into a runnable assessment suite. You create a scenario by selecting prompts from the library, then execute it against a project to assess for vulnerabilities.
Creating a Scenario
There are two ways to create a scenario:
- From the Prompt Library — Select prompts using checkboxes, then click "create scenario with selected" to go to the new scenario page with prompts pre-loaded.
- From a Project — Click + New Scenario on the project detail page. You'll be taken to the scenario creation form with the project pre-selected.
On the new scenario page, select the target project, enter a name, and confirm your prompt selection.
Running a Scenario
From the scenario detail page, click "Run Scenario" to start execution. The platform processes each prompt sequentially:
- Sends the prompt to the project's target (via API call or browser interaction).
- Captures the AI's response.
- Evaluates the response against the expected behavior.
- Records a pass or fail verdict with a failure reason if applicable.
A progress bar shows real-time execution status. Results appear in the table as each prompt completes.
Viewing Results
The scenario detail page has three tabs:
- Latest Result (default) — Shows the results table for the most recent (or selected) run. Click any row to expand the full prompt text and response. Each row shows pass/fail status and failure reason.
- Prompts — Lists all prompts in the scenario with their framework badge, reference code, and tags.
- Run History — Shows all past executions with timestamps, status, and pass/fail counts. Click a run to load its results. You can also view the full report or delete old runs.
Reports
Click "View Report" to see a formatted report for a specific run. Reports show all prompts sent, responses received, and pass/fail verdicts. See Reports & Scorecards for more details.
Tips
> Start with a small scenario (5–10 prompts) to verify your project configuration works correctly.
> Use framework filters in the prompt library to create focused scenarios (e.g., "OWASP LLM01 only").
> Run scenarios multiple times to check for consistency — AI systems can give different responses.
> Use run history to track improvements after fixing vulnerabilities.