# dsh-session-slm-router

[![npm version](https://img.shields.io/npm/v/dsh-session-slm-router)](https://www.npmjs.com/package/dsh-session-slm-router)
[![License](https://img.shields.io/badge/license-MIT-blue)](LICENSE)

Shadow-mode SLM router plugin for DeepSeek Harness. Per-turn weak/strong prediction via the `vertical-small-model` CLI; writes a shadow JSONL log **without changing the active session model** (S1 only).

> ⚠️ **S1** — default `shadow` mode: observe only, no model switching. `weak-only` mode (S3 grayscale) allows downgrade-to-weak switches but does not persist the user's original model.

[中文](README.md)

## Installation

```bash
# Via dsh plugin (recommended, auto-registers the bundle)
dsh plugin add dsh-session-slm-router
```

## Configuration

### Mode

| mode | Behavior |
|------|----------|
| `shadow` (default) | Predict + write shadow log, **no model switching** |
| `weak-only` (S3) | Downgrade to weak slot when predicted; `switch_to_strong` logged but not executed |
| `off` | Disabled |

### Full config example

```yaml
session-slm-router:
  mode: shadow
  weakOnlyMainOnly: true
  predictCmd: "python3 <path-to>/vertical-small-model/scripts/route_predict.py"
  predictModel: "<path-to>/vertical-small-model/data/eval/routing-v0/model-r1.json"
  timeoutMs: 250
  logPath: "slm-shadow/session-slm-shadow.jsonl"
  weakSlots:
    - { provider: <provider>, model: <model> }
  strongSlots:
    - { provider: <provider>, model: <model> }
```

- `predictCmd` / `predictModel` (required): path to the [vertical-small-model](https://github.com/NinjaSln-labs/vertical-small-model) CLI
- `weakSlots` / `strongSlots`: candidate model slots for routing

## Development

```bash
git clone https://github.com/NinjaSln-labs/dsh-session-slm-router.git
cd dsh-session-slm-router
npm install --legacy-peer-deps
npm run build
npm test
```

## License

MIT © 2026 ninjasln