How do I test PydanticAI agents without calling an LLM?

Use the TestModel class provided by PydanticAI, which allows you to simulate agent responses and tool calls deterministically without network requests.

Can I mock dependencies like databases in my agent tests?

Yes, the skill demonstrates how to use the agent.override context manager to swap production dependencies with mocks or stubs during test execution.

What are VCR cassettes in the context of AI testing?

VCR cassettes record actual LLM API responses to YAML files, allowing you to replay them in future test runs to ensure consistency and save on API costs.

How do inline snapshots help with AI testing?

They allow you to capture and assert against the current output of an agent, making it easy to detect changes in agent behavior over time without manual string matching.

PydanticAI Agent Testing

Name: PydanticAI Agent Testing
Author: existential-birds

byexistential-birds

•

セキュリティとテスト

Simplifies the creation of deterministic, recorded, and snapshot-based tests for PydanticAI agents and workflows.

This skill provides specialized patterns and guidance for testing PydanticAI agents without making live LLM calls. It enables developers to use deterministic TestModel instances for unit testing, record real API interactions via VCR cassettes for integration tests, and utilize inline snapshots for output verification. By facilitating mock dependencies and tool-call simulations, the skill ensures robust, reproducible, and cost-effective testing of complex AI agent logic within Python environments.

主な機能

01Recorded LLM interactions using VCR cassettes for offline integration testing

02Deterministic testing with TestModel and custom output overrides

03Inline snapshot assertion for automatic verification of structured agent outputs

0414 GitHub stars

05Custom logic implementation via FunctionModel for complex agent simulations

06Dependency injection support for mocking databases and external APIs during test runs

ユースケース

01Verifying tool-calling behavior and dependency injection in complex agent workflows

02Unit testing AI agent logic without incurring LLM costs or latency

03Creating regression tests for structured output agents using recorded VCR cassettes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add existential-birds/beagle pydantic-ai-testing

For use in Claude.ai and ChatGPT

Download Skill