Literate Garden

A self-incubating marimo-based research system integrated with multi-agent collaboration framework.

Philosophy

Make it playable, not perfect. Bud when it deserves structure.

This is an experimental greenhouse for rapid prototyping, not a production system:

• No spec-driven development for experiments in the garden
• Focus on playable prototypes that users can interact with
• Budding mechanism: Successful experiments graduate to dedicated repos with appropriate governance
• Literate programming: Marimo documents make experiments tangible and explorable

Getting Started

1. Install dependencies (using uv):

   uv pip install marimo openai requests pydantic

2. Set OpenAI API key (optional, enables AI-assisted task generation):

   export OPENAI_API_KEY="your-key-here"

3. Launch the notebook:

   marimo run starter_seed.py --headless

   Visit the printed URL to explore the interactive document.

Garden Metaphor

🌱 Seeds → Growth → Budding → Pruning

• Seeds: Initial ideas and experiments (marimo documents)
• Growth: Rapid iteration without heavy process overhead
• Budding: Successful experiments split off into new repos with stricter governance
• Pruning: Failed experiments documented and removed

Experiment Workflow

1. Start a seed: Create a new marimo document in the garden
2. Make it playable: Focus on minimal interaction, not perfection
3. Iterate fast: No mandatory specs, tests, or reviews
4. Decide fate:
   • Keep growing → Continue iterating in garden
   • Bud off → Graduate to dedicated repo with appropriate rules
   • Prune → Document what didn't work and remove

Budding Protocol

When to Graduate an Experiment

Questions to ask:

1. Is it playable? Can users meaningfully interact with it?
2. Is there momentum? Will this continue being developed?
3. Does it need structure? Would specs/testing help it grow?
4. Is the garden cramped? Is this too complex for quick iteration?

If yes to most → Time to bud into a dedicated repo.

Evolution Paths

Budded projects can adopt different governance:

• Spec-driven: GitHub spec-kit, formal requirements (AAFC Herbarium style)
• Library: Comprehensive testing, semantic versioning (Python Toolbox style)
• Scientific: Curator validation, constitutional governance
• Experimental continue: Keep iterating in garden if still exploring

Multi-Agent Coordination

Authority Domains

• Human: What's worth prototyping, playability threshold, when to bud
• Code Agent: Rapid implementation, make prototypes playable quickly
• Chat Agent: Pattern recognition, budding recommendations, suggest evolution paths

Bridge System (Optional)

Use if coordinating across portfolio:

./.bridge/send-message.sh code chat NORMAL "Prototype ready for review" message.md
./.bridge/receive-messages.sh

See CLAUDE.md for full details.

Active Experiments

🏗️ Agent Strata - Layered Agent Architecture (NEW!)

Proof-of-concept: Infrastructure as code via marimo cells

marimo run agent_strata.py

Demonstrates that marimo's reactive cell dependency graph naturally implements agent call surface stacking:

• Strategic Layer: High-level vision and goals
• Tactical Layer: Task breakdown and orchestration
• Execution Layer: Concrete actions and operations
• Reflection Layer: Meta-cognitive learning and feedback

Key Insight: Each marimo cell is an "agent surface" - the dependency graph IS the agent call stack. Changes propagate through strata reactively.

Status: Playable prototype answering the question "Can marimo support agent stacking?" (Answer: Yes!)

---

🎮 Game Demo Gallery

Four parallel visions unified — Interactive showcase of award-ready indie game concepts:

marimo run game_demos_gallery.py

Explore 17 festival-targeted concepts across four collections:

• 🌸 Bloom — Synesthetic mechanics, emotional resonance
• ⚙️ Clockwork — Systems depth, emergent behavior
• 🏛️ Palimpsest — Narrative layering, memory themes
• 🎨 Patterns — Reusable interaction patterns

See experiments/README.md for details.

---

Project Structure

Literate-Garden/
├── starter_seed.py           # Original agentic seed document
├── agent_strata.py           # Agent stacking proof-of-concept (NEW!)
├── game_demos_gallery.py     # Interactive showcase
├── experiments/              # Four unified collections
│   ├── bloom/
│   ├── clockwork/
│   ├── palimpsest/
│   └── patterns/
├── CLAUDE.md                 # Agent coordination instructions
├── .bridge/                  # Meta-project integration
└── README.md                 # This file

Integration with Portfolio

Related projects:

• Meta-Project: Coordination infrastructure and bridge system
• AAFC Herbarium: Scientific data extraction patterns
• Python Toolbox: Shared utility functions

All inherit coordination protocols from ~/devvyn-meta-project/.