agent-harness
What it is
A TypeScript test and evaluation harness that makes AI agents deterministic and testable using a record/replay mechanism for LLM interactions. It features a small payments-support agent to demonstrate trajectory evaluation, guardrails for safety, and structured observability traces.
The problem it solves
AI agents are non-deterministic and difficult to test; standard unit testing tools fail to handle LLM variability, and live testing is expensive, flaky, and requires API keys.
What you could use it for
- Testing AI agent behavior in CI without consuming API credits or facing non-determinism
- Detecting regressions in agent prompts or tool definitions via strict request fingerprinting
- Generating human-readable reports of agent traces including token costs and guardrail decisions
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.