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

# Personas

> A simulated person at the terminal. It drives a normal session on a scenario, so a harness change is measured before a person meets it.

## Run

```bash theme={"dark"}
hum persona run <name> [--scenario <s>] [--model <route>] [--session-model <route>]
hum persona runs <name>          # every run: verdict, level, cost, session
```

The run directory keeps the transcript, the persona's verdict, its answers to the scenario's checks, findings about the tool, and the cost to the person (messages, corrections, approvals, waiting, money).

## Files

`personas/<name>/persona.json` in the working directory or `~/.hum/personas/`: `role`, `brief` or `brief_file`, `model`.

Scenarios, one JSON each under `scenarios/`:

| Field                                     | Meaning                                                     |
| ----------------------------------------- | ----------------------------------------------------------- |
| `opening`                                 | First message, in the persona's words                       |
| `expectations`                            | What the persona holds the work to. The model never sees it |
| `world`                                   | Directory copied into the workspace                         |
| `checks`                                  | Yes/no questions answered at the end                        |
| `max_exchanges`, `max_minutes`, `max_usd` | Ceilings                                                    |

Hold the persona and session model fixed across harness versions. The difference is the harness.