Day 8: The Hybrid Decision
Formalizing the architecture: Skills for safety, Python for reasoning
Day 8: The Hybrid Decision
After a week of building, today was about deciding. How should ALOS actually be structured? What goes where? Why?
The Answer: Hybrid
The final architecture is intentionally hybrid:
Claude Code Skills handle:
- Interactive workflows (wrap-up, start-session)
- Safety-critical operations (Family Guardian with allergy hooks)
- Hot-reloadable behavior
- Declarative definitions (YAML frontmatter)
Python Agents handle:
- Complex reasoning chains
- API integrations (Obsidian MCP, calendar, email)
- Testable business logic
- Background processing
Neither is better. They complement each other.
The Cleanup
With the architecture decided, cleanup became obvious:
- Deleted 6 empty placeholder directories
- Fixed import paths (router → orchestrator)
- Updated README with complete hybrid documentation
- Added Claude Code Skills section to INTERFACES.md
- Got all 10 validation tests passing
What I Built
- Formalized hybrid architecture documentation
- Clean README explaining the system
- Updated INTERFACES.md with skills coverage
- Passing test suite (finally green across the board)
What I Learned
Architecture decisions are liberating. Before today, every new feature raised the question "where does this go?" Now there's a framework. Safety-critical and interactive → Skill. Complex reasoning and testable → Python.
Delete the placeholders. Empty directories aren't potential. They're clutter. If something doesn't exist yet, it shouldn't have a folder.
The Test
Tomorrow I'll test the architecture with a real invocation. /family-guardian Plan dinner with Jude should trigger safety hooks, check peanut allergy constraints, and provide a real answer.
If that works, the hybrid architecture isn't just theory. It's practice.