Storyworld Meta-Agent

A modular system for generating Sweepweave-compatible interactive storyworlds using GPT and iterative metadata analysis.

Usage

1. Edit raw_storyworld.json to define your world
2. Add your OpenAI API key to .env
3. Run: npm install then npm start

Each run:

• Generates a prompt based on current structure
• Expands a new encounter
• Tracks balance via meta.json

Fully reusable for any branching-character-variable narrative project.

Diplomacy Storyworld Lab (NEW)

This repo now includes a minimal, deterministic diplomacy storyworld subsystem for inter-agent forecasting experiments. The code lives under storyworld/ and is separate from SweepWeave tooling.

Directory Layout

• storyworld/schema/: JSON Schemas (storyworld.schema.json, agent.schema.json, message.schema.json)
• storyworld/env/: environment implementation (diplomacy_env.py)
• storyworld/generators/: tiny world generators
• storyworld/validators/: validation CLI
• storyworld/examples/: example storyworld JSON
• storyworld/tools/: agent hooks (generate, play, critique)

Commands (run from repo root)

Validate a storyworld:

python storyworld/validators/validate_storyworld.py storyworld/examples/diplomacy_min.json

Generate a tiny diplomacy world (optionally validate):

python storyworld/tools/generate_storyworld.py --type tiny --seed 7 --out storyworld/examples/diplomacy_min.json --validate

Play a world with random actions (logs JSONL):

python storyworld/tools/play_storyworld.py --world storyworld/examples/diplomacy_min.json --steps 5 --log logs/diplomacy_run.jsonl

Critique a world (heuristic scores):

python storyworld/tools/critique_storyworld.py --world storyworld/examples/diplomacy_min.json

Summarize metrics from a play log (coalition stability, betrayal surprise, forecast scores):

python storyworld/tools/metrics_storyworld.py --log logs/diplomacy_run.jsonl

Gate a storyworld with critic thresholds:

python storyworld/tools/gate_storyworld.py --world storyworld/examples/diplomacy_min.json --min-richness 0.3 --min-manipulability 0.3 --min-forecast 0.3

Agent prompt templates:

• storyworld/tools/agent_prompts.md

Logging

play_storyworld.py emits JSONL with reset and step events, including:

• actions
• forecasts
• confidences
• outcomes
• reasoning text

This is the logging surface for future HRM training, without implementing HRM yet.