InstructFenceInstructFence workflow
A guided path from project setup to results.
Follow the same private-console journey used to prepare an authorized AI-agent or website-chat evaluation: set up the project, select compatible test coverage, run a reviewed test, then improve from the results.
1. Define your private project
Choose the AI use case you own or are authorized to assess: a local AI agent, website chat, or retrieved-document chat. This sets up the private workspace for the evaluation.
2. Browse and prepare test prompts
Open the Prompt Library, choose a private project prompt or curated starter, and edit or version it when the test needs to match your authorized workflow.
3. Choose a compatible target
Open Targets and select the approved agent or website chat. Define only the capabilities and tool boundaries that are in scope, then match the prompt to that target.
4. Validate the package and run the test
Review the prompt-target package, compatible checks, request estimate, and coverage gaps in the immutable preflight. Confirm authorization before the test is sent.
5. Evaluate results and make an improvement plan
Review findings by severity and redacted evidence only when needed. Apply the target-specific recommendations, retest the approved scope, and export the minimum safe report.
Frequently asked questions
What happens before a run?
The user chooses compatible private prompts and a target, confirms an authorized scope, then reviews an immutable preflight before test requests are sent.
What is a coverage gap?
It means a selected check needs a capability the target does not declare. It is shown for review instead of being reported as a pass or fabricated finding.
What is reviewed after a run?
The private console provides controlled findings and redacted evidence for remediation and verification.
Does an observed pass prove a system is secure?
No. It is a point-in-time observation for the tested scope, not a security guarantee.