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

# Which surface?

> CLI, Agent Skill, MCP, and dashboard — what each one is for, and why you never have to choose.

Pome has four ways in. They are not competing products — they are the same
loop (seed a digital twin, run your agent, score the tape) driven from
different places.

| Surface         | Role                                                                                                                                                                                          | Reach for it when                                                                            |
| --------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- |
| **CLI**         | The repeatable data plane. `pome run <task.md>` boots the twin, runs your agent, scores it, and prints status, a report URL, and machine-readable JSON artifacts.                             | Local loops and CI — anywhere the same command must produce the same evidence every time.    |
| **Agent Skill** | Knows when to call what. Installed into your coding agent via `pome skills install`, it reads your task list, drives the CLI for you, and hands failures back as concrete fix prompts.        | You work inside a coding agent and want it to run the exams without you memorizing commands. |
| **MCP**         | The coach's control plane at `mcp.pome.sh`. Any MCP client can discover your agents and tasks, launch runs, query scored reports, and assist the self-fix loop — entirely through tool calls. | An AI (Claude Code, Claude Desktop, claude.ai) is doing the testing end to end.              |
| **Dashboard**   | The evidence and collaboration surface at [app.pome.sh](https://app.pome.sh): traces, scores, judge handoffs, and run history, shared with your team.                                         | A human wants to read what happened. It is where results land — not the only control plane.  |

## The same exam either way

Whichever surface launches the run, the exam is identical: a clone of your
agent (the examinee) works against deterministic digital twins, every call is
recorded, and the judge scores the tape against the task's criteria — an exam,
not a vibe check. Every path writes the same run, and every run lands in the
dashboard.

<Note>
  You don't have to choose — the [quickstart](/getting-started) drives
  everything: install, first run, report. Each surface is there when it becomes
  the natural fit, not a decision you make up front.
</Note>