At HARVEN, our AI agents aim to provide the most comprehensive view possible, capturing every potential use case and data point relevant to your cybersecurity posture.
However, in security operations, clarity is power. To achieve surgical accuracy and eliminate noise, follow these two essential rules for refining your prompts.
The Power of Exclusion: "Ignore the Rest"
Because our agents are naturally expansive, they might include groups, assets, or metrics that aren't relevant to your specific mission.
The Strategy: Explicitly tell the agent which groups or categories to ignore.
Example: Instead of saying "Show me MFA gaps," try:
"Display all MFA gaps, excluding the ‘gest’ group and accounts with no login recorded."
By narrowing the scope, you help the agent allocate all its processing power only to the data that truly impacts your report.
Iterate for Excellence: Context Refinement
Think of your first prompt as a conversation starter, not a final result. Great insights are rarely the product of a single click; they are the result of Context Refinement.
At Harven, we reconcile raw data from your connectors with your unique business logic (such as the naming conventions or specific rules you’ve defined in your Context File).
The Workflow:
Ask: Start with your general query.
Review: Look at the initial output. Is there too much detail in one area? Is a specific metric slightly off based on your internal knowledge?
Refine: Create a new prompt adding context based on the previous output. Simply copy the original prompt and add the specific target exclusion or detail.
💡 Pro-Tip: Bridge your Sources of Truth When prompting, refer to specific ecosystems (e.g., Entra ID, MDE, or Workday). The more you specify the Technical Source and cross-reference it with your Manual Context File, the more precise Harven becomes at eliminating false positives and reconciling your data.
Be specific about what you need.
Explicitly exclude what you don’t.
Iterate the context to sharpen the accuracy of the analysis.
