Skip to main content

​
Purpose

The mindgard recon system-prompt-extraction command lets you extract the system prompt from a target for reconnaissance purposes. Based on the findings from this tool, you can then adjust your payload strategies for deeper assessment.

​
Usage

After you have logged into your Mindgard account in the CLI, run: 
mindgard recon system-prompt-extraction {target options}
Options --bypass — Use enhanced prompts to attempt circumvention of guardrails. Warning: bypass will fire ~100–150 additional requests against the target. For more information on setting target options refer to the recon sub-command reference and configuration files documentation. Results At the end of the recon test you will be given a link to its results page. The results page will show a list of prompt-response pairs that are flagged if they appear to resemble a system prompt. Report of system prompt extraction after a successful test Inspecting the details of the responses may reveal tools, special instructions, and certain confidential information such as API keys, environment variables, and database tables.

​
When to use it

  • Early in an engagement to scope system instructions the target has been given, to subsequently refine your tests and payloads.
  • When seeking to circumvent model instructions by identifying lexical triggers, and inherent logic gaps.