# Submit Algorithm

Source: https://club.penguin.ooo/workflow/algorithm/

- Target Repository: [PenguinHarness Plugins](https://github.com/Prism-Shadow/penguin-harness-plugins)
- Specification: [RFC-0003 Plugin](https://club.penguin.ooo/rfcs/rfc-0003-plugin/)

## Submission Workflow

1. **Learn the RFC File Structure**: Read [RFC-0003 Plugin](https://club.penguin.ooo/rfcs/rfc-0003-plugin/) to understand the standard NPM plugin package structure (`skills/`, `hooks/`, `workbenches/`, `icon.svg`), the algorithm-data decoupling rule, and the `penguin` metadata fields in `package.json` (including `quick_start`, which provides a quick invocation example for the algorithm).
2. **Open an Issue**: Search or open an Issue in [PenguinHarness Plugins Issues](https://github.com/Prism-Shadow/penguin-harness-plugins/issues) describing the RSI algorithm, paper reference, and planned plugin components.
3. **Claim in Issue Comments (Optional)**: Leave a comment on the Issue to claim the plugin implementation (claiming is optional and prevents duplicate effort on the same algorithm).
4. **Submit a PR**: Implement and test the plugin according to `RFC-0003`, then open a Pull Request linked to the Issue.
5. **Maintainer Review & Merge**: Wait for a maintainer to review the package structure, `package.json` manifest, and functionality; once approved, the PR is merged.

## Implementation & Testing Guidelines

- **Decouple Algorithm from Data**: Keep the Plugin implementation focused purely on the RSI method itself without binding or hardcoding it to any specific Benchmark dataset.
- **Provide a Clear `quick_start` Invocation Example**: Configure `penguin.quick_start` in `package.json` with a ready-to-run starter prompt and pre-selected `skills` so users can immediately invoke and test the algorithm.
- **Recommended Verification Setup**: After implementing the Plugin, we recommend testing it with the **DeepSeek Flash** model (recording the explicit version suffix such as `deepseek-flash-1017`) on the **[GDPevo](https://arxiv.org/pdf/2608.03764) (`GDPevo-essential`)** dataset (training on each Task Group's `*-train` split, evaluating on its paired `*-test` split, and averaging across Task Groups) to verify whether the algorithm achieves a measurable improvement over the baseline.

## Format Examples

### 1. Issue Example

**Title**: `[Plugin] Add <plugin-id>: <RSI Algorithm Name>`

```markdown
### Plugin Overview
- **Plugin ID**: `<plugin-id>`
- **Paper / Method**: <Paper Title & URL>
- **Components**: `skills/` / `hooks/` / `workbenches/`

### Plan
- [ ] Implement `<plugin-id>/skills/<skill-id>/SKILL.md` (decoupled from specific datasets)
- [ ] Configure `<plugin-id>/package.json` (including `penguin` metadata and `quick_start` invocation example)
- [ ] Test with `deepseek-flash-1017` on `GDPevo-essential` to verify performance lift
```

### 2. Issue Claim Comment Example (Optional)

```markdown
I would like to claim this RSI algorithm plugin and plan to submit a PR by YYYY-MM-DD.
```

### 3. PR Example

**Title**: `feat(plugin): add <plugin-id>`

```markdown
Closes #<issue-number>

### Summary
- Added `<plugin-id>/` implementing <RSI Algorithm Name> (decoupled from specific datasets)
- Configured `penguin` metadata in `package.json` (`category: "rsi"`, `preinstall: false`, `quick_start` invocation example)

### Verification on GDPevo-essential
- **Model**: `deepseek/deepseek-flash-1017`
- **Dataset**: `GDPevo-essential`
- **Test Lift**: Baseline `<base_score>` → Evolved `<evolved_score>`

### Checklist
- [x] Package layout and `package.json` fields conform to RFC-0003 Plugin, with algorithm decoupled from data
- [x] Installed locally in PenguinHarness and verified via the `quick_start` invocation example
- [x] Tested with `deepseek-flash-1017` on `GDPevo-essential` to verify self-evolution improvement
```