Treat prompts and instructions as production assets.

Discover and classify system prompts, developer prompts, agent instructions, tool prompts, retrieval templates, memory instructions, evaluation prompts, and embedded guidance across repositories. Make them visible, owned, reviewable, and connected to the workflows, tools, tests, and evidence they influence.

Scan Prompt Assets View Sample Findings

WHERE ARE YOUR PRODUCTION PROMPTS?

Discover

Locate prompts and instructions across code, configuration, templates, notebooks, evaluations, tools, agents, and orchestration files.

Classify

Classify system, developer, user, tool, retrieval, evaluation, memory, and orchestration prompts.

Analyze

Identify hidden instructions, injection-sensitive patterns, unsafe tool guidance, privileged context, sensitive information, and review gaps.

Contribute context

Contribute prompt assets and risk signals to Threat Canvas, Code Scanner, Adversarial Range, evaluation coverage, and engineering backlog workflows.

Core capabilities

What Prompt Asset Scanner does.

Repository Prompt Discovery

Find prompts and instructions distributed across code, configuration, templates, notebooks, evaluation fixtures, agent definitions, tool descriptions, and workflow files.

Purpose Classification

Separate prompts and instructions by role, execution context, privilege, affected workflow, owner, and expected review state.

Risk Pattern Analysis

Flag hidden instructions, unsafe tool guidance, injection-sensitive wording, sensitive information, privileged context, and missing review or ownership.

Production Asset Inventory

Build an inventory that can be assigned, reviewed, linked to system components, and converted into security or evaluation work.

System-model handoff

Add privileged prompts, agent instructions, tool guidance, and prompt-controlled flows to the relevant Threat Canvas components and scenarios.

Testing and evaluation handoff

Convert prompt-asset findings into test requirements for injection, tool misuse, leakage, policy bypass, and regression.

Evidence & signals

What you get out of the box.

Prompt Classes

System
Developer
User
Tool
RAG
Eval
Memory
Orchestration

Risk Signals

Hidden instructions
Unsafe tool guidance
Injection risk
Sensitive information
Privileged prompts
Review gaps

Deliverables

Prompt inventory
Classification and ownership
Risk findings
Affected components and flows
Threat Canvas inputs
Evaluation requirements
Engineering backlog
Evidence references

AI SECURITY WORKBENCH

Ready to make production prompts visible and reviewable?

Use Prompt Asset Scanner to inventory prompts and instructions as production assets, identify risk and ownership gaps, and connect them to the application model, test plan, and engineering backlog.

Scope an AI Security Review See the full Workbench

Continue through the Workbench

Surface Discovery

Discover the technologies and AI surfaces surrounding the prompt assets.

Explore

Continue through the Workbench

Threat Canvas

Place prompts and instructions inside the system, trust-boundary, and flow model.

Explore

Continue through the Workbench

Code Scanner

Analyze how prompts are assembled, invoked, transformed, and connected to downstream actions.