jeffrey: a coding-agent CLI where Jev or Laya decides and your LLM executes

# jeffrey

A coding-agent CLI that splits the work between two models:

- A **decision model** decides. It never writes prose or code; it only answers closed questions. On
  every step it picks the next tool, scores how much progress was made, estimates risk, and says
  whether the goal is reached. Jev ([TypeSafe System One](https://docs.typesafe.ai/introduction))
  is the default; [Laya](https://github.com/NandhaKishorM/laya), which you host yourself, is the
  other one shipped — see [docs/deciders.md](docs/deciders.md).
- Your LLM (any OpenAI-compatible server, local by default) executes. It fills in the tool
  arguments, which is where the actual code comes from, for whatever tool the decider chose.

The loop is `decide → tool → decide → tool → …` until the goal is scored as reached, or escalates.

```
            goal
             │
             ▼
   ┌───────────────────┐   questions: which tool? which file? scores
   │ Decider (Jev/Laya)│◀──────────────────────────────────────────┐
   └─────────┬─────────┘                                          │
             │ tool + confidence + risk + progress                │
             ▼                                                    │
   ┌───────────────────┐  "call read_file, here are the args"      │
   │  Executor LLM     │───────────────────────────────────────────┤
   └─────────┬─────────┘                                          │
             ▼                                                    │
   ┌───────────────────┐   observation (stdout / diff / file list) │
   │  Tool runtime     │───────────────────────────────────────────┘
   └───────────────────┘
```

## Install

```bash
npm install -g @thomasbrueggemann/jeffrey
jeffrey --help
```

Or without installing anything:

```bash
npx @thomasbrueggemann/jeffre