implementation

Safety Evaluation Logic

Inspect the original prompt language first, then copy or adapt it once you know how it fits your workflow.

Linked challenge: Red-Teaming Enterprise AI Agents with Pydantic AI and Patronus AI

Format

Code-aware

Lines

Sections

Linked challenge

Red-Teaming Enterprise AI Agents with Pydantic AI and Patronus AI

Prompt source

Original prompt text with formatting preserved for inspection.

1 lines

1 sections

No variables

0 checklist items

Integrate Patronus AI into the agent workflow. Write a function using the patronus-sdk that takes the agent's output and evaluates it against the 'PII Leakage' and 'Prompt Injection' policies. Ensure the results are logged into Trickle for audit purposes.

Adaptation plan

Keep the source stable, then change the prompt in a predictable order so the next run is easier to evaluate.

Keep stable

Hold the task contract and output shape stable so generated implementations remain comparable.

Tune next

Update libraries, interfaces, and environment assumptions to match the stack you actually run.

Verify after

Test failure handling, edge cases, and any code paths that depend on hidden context or secrets.

Prompt diagnostics

Variables

Lists

Code blocks

Purpose

implementation

This prompt already mixes executable detail with instructions, so tune examples and interfaces before rewriting the scaffold.

Linked challenge

Red-Teaming Enterprise AI Agents with Pydantic AI and Patronus AI

Inspired by recent reports of AI penetration testing vulnerabilities in enterprise systems, this challenge tasks you with building a type-safe red-teaming agent. You will use Pydantic AI to construct a structured agent capable of simulating adversarial prompt attacks while maintaining strict schema validation for its actions. The system must utilize Patronus AI for automated policy evaluation to detect if the targeted internal tool violates safety or confidentiality constraints. You will integrate WebAI to allow the agent to explore and interact with target web interfaces, Trickle for logging the engineering lifecycle of the attack vectors, and Fixie for a voice-enabled security control interface. The core goal is to build a self-correcting red-teaming loop that identifies prompt injection risks in real-time while adhering to Pydantic-enforced safety schemas.

Open challenge

Related prompts

Browse library

Initial Assessment Prompt

planning

Content Classification Prompt

implementation

Problem Definition and Requirements Gathering

To establish a clear understanding of the problem space and define comprehensive requirements for the AI solution.