> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usefused.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Testing and compilation

> Run unit tests, smoke tests, evals, CI checks, and compile standalone artifacts.

Harnest owns the environment and compiled application used by every test lane. Authored tests exercise the selected ADK or LangGraph target without importing Harnest or manually loading build artifacts.

| Lane    | Command                          |         Network | Best for                           |
| ------- | -------------------------------- | --------------: | ---------------------------------- |
| Unit    | `harnest test AGENT_DIR`         |              No | Agent logic and local tools        |
| Smoke   | `harnest test AGENT_DIR --smoke` |         Allowed | Live models, MCP, and API journeys |
| Evals   | `harnest test AGENT_DIR --evals` | Depends on eval | Quality and tool trajectories      |
| Compile | `harnest compile AGENT_DIR`      |              No | Reproducible standalone artifact   |

## Unit tests

Run the default offline lane:

```bash theme={null}
harnest test AGENT_DIR
```

Harnest compiles the project, then collects `tests/unit/test_*.py`.

| Fixture | Value                        |
| ------- | ---------------------------- |
| `agent` | Selected framework target    |
| `tools` | Read-only map of local tools |

Unit tests should not call models, MCP servers, or other networks. This is a convention, not an OS sandbox.

## Smoke tests

Live runtime checks belong in `tests/smoke/test_*.py` and run only when selected:

```bash theme={null}
harnest test AGENT_DIR --smoke
```

The unit lane runs first. Smoke tests add a FastAPI `client` and a `smoke` fixture.

| Method               | Returns                   |
| -------------------- | ------------------------- |
| `smoke.respond(...)` | One neutral JSON response |
| `smoke.stream(...)`  | Ordered SSE event objects |

Smoke tests may spend time or money and use real credentials. Run them deliberately.

## Evaluations

Run eval assets after the Python tests:

```bash theme={null}
harnest test AGENT_DIR --evals
harnest test AGENT_DIR --smoke --evals
```

ADK projects can place official `EvalSet` JSON files in `evals/`. The default `business` trajectory requires business calls in order but allows helper calls. Require the exact sequence with:

```bash theme={null}
harnest test AGENT_DIR --evals --eval-trajectory strict
```

ADK eval files do not run for LangGraph. Put portable evaluations in unit or smoke tests.

## Compile an artifact

Compile validated source into a standalone runtime directory:

```bash theme={null}
harnest compile AGENT_DIR --output .harnest/my-agent
```

| Artifact file           | Purpose                             |
| ----------------------- | ----------------------------------- |
| Preserved source        | Your agent code                     |
| Generated adapters      | Selected framework integration      |
| `harnest-manifest.json` | Versions, mode, and source identity |
| `server.yaml`           | Mutable runtime settings            |
| `harnest-agent`         | Launcher                            |

Compiled directories are reproducible output. Do not edit or commit them.

## CI

A minimal offline CI sequence is:

```bash theme={null}
harnest env sync AGENT_DIR --frozen
harnest test AGENT_DIR
harnest compile AGENT_DIR --output .harnest/my-agent
```

Add smoke or eval lanes only when CI has the required services and credentials. For Harnest source changes, follow [Development standards](/harnest/reference/development-standards).