# AI Harness Testing

Test harness behavior with deterministic tools, mocked models, and regression cases.

---
Canonical: /harness/testing/
Source: web/src/data/harness-markdown.ts
Format: Markdown for agents
---

Test agents like production application code.

Use mocked models for deterministic paths and evaluation datasets for behavioral quality. Keep tool tests separate from model-behavior tests.

## What To Test

- Tool schema validation.
- Authorization and data minimization.
- Expected tool selection.
- Failure and retry behavior.
- Prompt-injection resistance.
- Output shape and refusal behavior for unsafe requests.