/orion-init - Initialize Orion Research Project

Initialize a new Orion research project with the directory structure and skills templates.

Usage

/orion-init "<project-name>"

Examples

/orion-init "video-understanding"
/orion-init "vlm-improvement"

Process

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

Step 1: Create Project Structure

mkdir -p research/{skills/{literature,domain,learned,benchmarks},papers,codebase,logs,results}

Step 2: Initialize Skill Templates

Create the Tree of Skills structure:

research/skills/literature/_overview.md

# Literature Overview

Papers analyzed and key insights extracted.

## Papers
| ID | Title | Key Method | Status |
|----|-------|------------|--------|

## Methodologies Extracted
[To be populated by /literature-review]

## Experiments & Benchmarks
[To be populated by /literature-review]

## Metrics
[To be populated by /literature-review]

---
*Run /literature-review <folder> to populate*

research/skills/domain/_overview.md

# Domain Knowledge

User-provided expertise and constraints.

## Background
- Domain: [To be filled]
- Expertise: [To be filled]

## Constraints
- Compute: [To be filled]
- Time: [To be filled]
- Data: [To be filled]

## Domain Intuitions
[Ideas the user believes might work]

research/skills/learned/_overview.md

# Learned Skills

Auto-generated after each hypothesis test.

## Summary
| Hypothesis | Result | Key Insight |
|------------|--------|-------------|

## Patterns
### What Works
### What Doesn't

research/skills/benchmarks/_overview.md

# Benchmarks

Benchmark datasets and evaluation protocols.

## Available Benchmarks
| Name | Task | Metrics | Status |
|------|------|---------|--------|

## Download Status
[To be populated by /benchmark-setup]

## Evaluation Protocols
[To be populated by /benchmark-setup]

Step 3: Create orion.json

research/orion.json

{
  "project": "<project-name>",
  "status": "initialized",
  "phases": {
    "init": "complete",
    "literature_review": "pending",
    "benchmark_setup": "pending",
    "hypothesis_generation": "pending",
    "experimentation": "pending"
  },
  "papers_folder": null,
  "benchmarks": [],
  "hypotheses": [],
  "current_hypothesis": null,
  "best_result": null
}

Step 4: Output Summary

Orion project initialized: <project-name>

Directory structure:
  research/
  ├── orion.json           # Project state
  ├── skills/              # Tree of Skills
  │   ├── literature/      # Paper insights
  │   ├── domain/          # Your expertise
  │   ├── learned/         # Experiment learnings
  │   └── benchmarks/      # Benchmark info
  ├── papers/              # Place papers here
  ├── codebase/            # Cloned repos
  ├── logs/                # Experiment logs
  └── results/             # Output files

Next steps:
  1. Place your paper PDFs in research/papers/
  2. Run: /literature-review research/papers
  3. View knowledge: /knowledge

Tools Used

• Bash: Create directories
• Write: Create template files and orion.json

Note

This command only initializes the structure. Use subsequent commands to build knowledge:

• /literature-review - Extract knowledge from papers
• /knowledge - View accumulated knowledge
• /benchmark-setup - Setup evaluation benchmarks
• /hypothesis-generation - Generate and test hypotheses