What is Prompt Studio?
Prompt Studio is an interactive environment for testing and iterating on the prompts that power your Alert Agents and Digest Agents. Test prompts against real data, compare versions, and save improvements back to your agents without recreating them.Key Capabilities
- Test investigation runbooks against real alert payloads to see how TierZero would investigate
- Test digest section prompts against real data to preview report output
- Compare prompt versions side-by-side to evaluate improvements
- Save refined prompts back to agents without recreating them
How to Use
Step 1: Navigate to Prompt Studio
Go to Prompt Studio in the sidebar.Step 2: Select Prompt Type
Choose what you want to test:- Alert Investigation Runbook: The instructions TierZero follows when investigating an alert
- Impact & Severity Analysis: How TierZero assesses alert impact and severity
- Digest Section: The prompt for a specific section of a digest report
Step 3: Provide Test Input
Enter a real alert payload or relevant context. Using real data from alerts that produced unsatisfactory results is the best way to improve prompt quality.Step 4: Run and Review
Run the test and review the output. TierZero executes the prompt against your test input using the same tools and data sources it would use in production.Step 5: Iterate
Make changes to the prompt and re-run to compare. The side-by-side view shows how different prompt versions produce different results.Step 6: Save
When satisfied with the output, save the improved prompt back to the agent. The update takes effect immediately for future runs.Best Practices
1. Start with Real Alerts- Use an actual alert that produced unsatisfactory results as your test input
- This grounds your iteration in a real problem
- Change one thing at a time and compare outputs
- Large rewrites make it hard to tell what improved the results
- A prompt that works well for one alert type might fail on another
- Test across different categories to ensure consistency
- Once a prompt version produces consistently good results across diverse inputs, save it back to the agent