/orion-setup - Setup Research Codebase

Clone and configure the research codebase. Verify baseline reproduction.

Usage

/orion-setup <github-url> [--branch <branch>] [--verify-baseline]

Examples

/orion-setup "https://github.com/author/paper-code"
/orion-setup "https://github.com/author/paper-code" --verify-baseline
/orion-setup ./existing-codebase --local

Process

IMPORTANT: This command MUST use Plan Mode. Create a plan first, get user approval, then execute.

When user invokes /orion-setup, follow this process:

Phase 1: Planning (Plan Mode)

Step 1: Analyze Repository

Use EnterPlanMode, then if GitHub URL provided:

# Check repo structure via GitHub API or clone

If local path:

ls -la <path>
cat <path>/README.md
cat <path>/requirements.txt  # or environment.yml

Step 2: Detect Environment

Look for environment indicators:

• environment.yml → Conda
• requirements.txt → pip
• pyproject.toml → pip/poetry
• Dockerfile → Docker
• setup.py → pip install

Step 3: Create Setup Plan

## Codebase Setup Plan

### Repository
- URL: [GitHub URL]
- Branch: [branch]
- Will clone to: research/codebase/

### Environment Detection
- Type: [Conda/pip/Docker]
- File: [environment.yml / requirements.txt]

### Setup Steps
1. Clone repository
2. Create [Conda env / venv]
3. Install dependencies
4. Check for data/weights downloads
5. Verify baseline (if --verify-baseline)

### Potential Issues
- [Any detected issues from README]
- [GPU requirements]
- [Data download requirements]

### User Decisions
- Environment name: [suggestion]
- GPU required: [Yes/No]

Step 4: User Approval

Use ExitPlanMode to get approval.

Phase 2: Execution (After Approval)

Step 5: Clone Repository

git clone <url> research/codebase
cd research/codebase
git checkout <branch>

Record commit hash for reproducibility.

Step 6: Setup Environment

For Conda:

conda env create -f environment.yml -n orion-research
# Or
conda create -n orion-research python=3.10
conda activate orion-research
pip install -r requirements.txt

For pip/venv:

python -m venv research/venv
source research/venv/bin/activate
pip install -r requirements.txt

Step 7: Check Data/Weights

Look for download requirements:

grep -r "download" README.md
ls scripts/download*.sh

If found, inform user:

Data/weights download may be required:
- [Script or instructions found]

Download now? [Y/N]

Step 8: Verify Baseline (if --verify-baseline)

Find and run evaluation:

# Find eval script
ls *.py | grep -E "(eval|test|infer)"

# Run baseline
python eval.py --config baseline.yaml --subset

Record baseline result:

{
  "baseline": {
    "method": "[method name]",
    "score": [score],
    "metric": "[metric]",
    "commit": "[hash]",
    "date": "[timestamp]"
  }
}

Step 9: Setup Branches

git checkout -b orion/main  # Our stable branch

Step 10: Create Codebase Skill

Create research/skills/codebase.md:

# Codebase: [Repo Name]

## Repository
- URL: [URL]
- Commit: [hash]
- Branch: orion/main

## Structure

codebase/ ├── [key directories] └── [key files]

## Key Files
| File | Purpose |
|------|---------|
| [file] | [purpose] |

## How to Run

### Inference
```bash
python eval.py --config config.yaml

Training (if applicable)

python train.py --config config.yaml

Configuration

Key config options:

• [option 1]: [what it does]
• [option 2]: [what it does]

Baseline Result

• Score: [score]
• Metric: [metric]
• Command: [exact command used]

#### Step 11: Update orion.json

```json
{
  "codebase": {
    "url": "[url]",
    "path": "research/codebase",
    "commit": "[hash]",
    "branch": "orion/main"
  },
  "baseline": {
    "score": [score],
    "metric": "[metric]"
  }
}

Step 12: Summary

Codebase Setup Complete!

Repository: [URL]
Location: research/codebase/
Commit: [hash]
Environment: [Conda/venv] - [name]

Baseline: [score] ([metric])

Branch structure:
- orion/main (stable baseline)

Next steps:
1. Generate hypotheses: /hypothesis-generation
2. Or view knowledge: /knowledge

Tools Used

• Bash: Git operations, environment setup
• Read: Check files
• Write: Create skill files
• EnterPlanMode/ExitPlanMode: Planning workflow

When to Use

Run /orion-setup after:

• /literature-review (understand what code to get)
• /benchmark-setup (know what to evaluate on)

Run before:

• /hypothesis-generation (need codebase to implement)