Imported from ActiveInferenceInstitute/Start (
.aii/SKILL.md). Install upstream withnpx skills add ActiveInferenceInstitute/Start --skill .aii. Copyright stays with the author.
START — Active Inference Curriculum Generator
Overview
START generates professional-grade Active Inference curricula using:
- Perplexity API for real-time domain/entity research
- OpenRouter API for LLM-based content generation and translation
- Matplotlib/Seaborn for visualization
- Multilingual output (11+ languages with script mapping)
Quick Start
git clone https://github.com/ActiveInferenceInstitute/Start.git
cd Start
# Install dependencies
uv sync --all-extras --dev
# Run tests
uv run pytest -q
# Run the curriculum generation pipeline (interactive)
uv run python learning/curriculum_creation/generate_custom_curriculum.py
# Or use the browser GUI
uv run python learning/curriculum_creation/generate_curriculum_gui.py
# Or run individual pipeline stages:
uv run python learning/curriculum_creation/1_Research_Domain.py
uv run python learning/curriculum_creation/1_Research_Entity.py
uv run python learning/curriculum_creation/2_Write_Introduction.py
uv run python learning/curriculum_creation/3_Introduction_Visualizations.py
uv run python learning/curriculum_creation/4_Translate_Introductions.py
Architecture
src/common/ — Shared utilities (io, paths, config, prompts, logging)
src/config/ — Language configuration
src/perplexity/— LLM API clients (Perplexity + OpenRouter), domain/entity/curriculum/translation
src/repos/ — Repository cloning for Active Inference ecosystem
src/system/ — Environment setup, dependency checking, reporting
src/terminal/ — Matrix-themed CLI (animations, colors, menu)
src/visualization/ — Curriculum metrics visualization
Pipeline scripts under learning/curriculum_creation/ are thin orchestrators
that import from src/ modules.
Configuration
data/config/domains.yaml— Domain definitions (keywords, categories, priorities)data/config/entities.yaml— Entity/audience profilesdata/config/languages.yaml— Translation targets and script mappingsdata/prompts/*.md— Prompt templates with{{variable}}substitution
Key Facts
- Test suite — all tests use real local protocols and temporary data
- Provider tests — local HTTP completion servers exercise the production client
- Standalone — not dependent on the template infrastructure
- Path structure: outputs under
data/ - Documentation: 23+ docs pages, AGENTS.md per subpackage