# LlmBench Profiles, Instructions, and AI Skills

> Understand LlmBench profile and instruction tooling, its relationship to trvny/.ai, and the boundary for future portable configuration integration.

LlmBench includes profile, instruction, YAML, and skills/docs tooling. The separate `trvny/.ai` repository remains the canonical portable configuration project, but LlmBench does not currently perform automatic runtime synchronization with it.

## Relationship to `.ai`

The `trvny/.ai` repository remains the canonical portable AI configuration core. LlmBench contains the multiplatform client code; `trvny/.ai` contains the reusable AI configuration you can bring into any compatible client.

```text
trvny/.ai             reusable profiles / instructions / skills
      │
      └── optional consumption
              │
              ▼
twojstar/llmbench      shared KMP core + platform clients
```

<Warning>
  Do not vendor a second copy of `.ai` into LlmBench. If runtime integration becomes useful, consume a pinned or exported representation with an explicit boundary rather than duplicating content.
</Warning>

## What profiles and instructions contain

The `.ai` project defines portable, reusable AI configuration that can include:

<CardGroup>
  <Card title="Profiles" icon="user">
    Named configurations that describe how a model or assistant should behave, including system instructions and parameter defaults.
  </Card>

  <Card title="Instructions" icon="file-lines">
    Specific prompts or prompt templates that can be reused across conversations or providers.
  </Card>

  <Card title="Skills" icon="wrench">
    Domain-specific capabilities and workflows defined as portable configuration.
  </Card>
</CardGroup>

## Runtime integration status

Automatic runtime consumption of `trvny/.ai` is **not implemented yet**. The current app owns its local profile/instruction state and tooling. If cross-repository runtime integration is added later, it should consume an explicitly pinned or exported representation so releases remain reproducible and the boundary stays auditable.

## Skills and docs browser

The first LlmBench prototype includes a small skills/docs browser that lets you:

- Browse the profiles and instructions available to the app
- View YAML definitions
- Apply configurations to conversations

<Note>
  This browser is part of the early prototype and will evolve as the `.ai` integration matures.
</Note>