little-coder: 소형 LLM에 맞춰 튜닝된 로컬 모델 코딩 에이전트 스캐폴드

I can't run commands or install packages for you, but here's exactly how to do each step yourself.

## 1. Install little-coder globally

```bash
npm install -g little-coder
```

Requires Node 22.19+. Verify with `node --version` first.

## 2. List available models

From your repo root:

```bash
little-coder --list-models
```

This prints the models little-coder knows about. Look for a local llama.cpp entry (often named something like `llama.cpp/Qwen3.6-35B-A3B` or a custom alias you configured).

## 3. Launch against your project

**If you have a local llama.cpp Qwen3.6-35B-A3B server running** (e.g. `llama-server` on `http://localhost:8080`), point little-coder at it:

```bash
little-coder --model llama.cpp/Qwen3.6-35B-A3B
```

(Use the exact identifier shown by `--list-models`. If your llama.cpp endpoint isn't auto-detected, you may need to set the base URL via config or an env var such as `LITTLE_CODER_BASE_URL=http://localhost:8080/v1` — check `little-coder --help` for the flag name in your version.)

**If you don't have one served**, tell me which model from the `--list-models` output you want to use, and I'll tailor the rest of the workflow to it.

## 4. Plan Mode workflow

Once launched, in the interactive session:

1. Switch to **Plan Mode** (usually `/plan` or a mode toggle — check `/help`).
2. Ask it to research the codebase using sub-coders, e.g.:
   > Use sub-coders to map the repo: entry points, main modules, test setup, and any build/config files. Summarize findings; do not edit anything.
3. It will ask you 1–3 clarifying questions with suggested answers — answer them.
4. It writes the plan **in chat only**, no file edits.

## 5. Save the approved plan and implement

After you approve the plan in chat:

```
Save the approved plan to .pi/approved-plan.md
```

Then:

```
/implement
```

This opens a fresh session seeded with `.pi/approved-plan.md` and begins the work.

## 6. Confirm bundled extensions

```
/extensions
```

This lists which bundled extensions loaded. Review the output to confirm the default set is active.

## 7. Add your own extensions without changing defaults

Drop your extension files into:

```
~/.config/little-coder/extensions/
```

little-coder loads user extensions from this directory **in addition to** the bundled set — the defaults are untouched. Typical layout:

```
~/.config/little-coder/extensions/
└── my-extension/
    ├── manifest.json   (or package.json, depending on version)
    └── index.js        (or index.ts)
```

Restart the session (or re-run `/extensions`) to confirm your extension appears alongside the bundled ones. If it doesn't load, check `little-coder --help` or the docs for the exact manifest schema your version expects.

---

If you paste the output of `little-coder --list-models` and `/extensions`, I can help you pick the right model identifier and verify your extension is being picked up.