This V1 is intentionally minimal. The architecture is designed to support more over time. The core principle: the repo is a substrate — a pile of markdown — and tools that act on it live outside. This keeps the system portable and lock-in free. See Design Decisions for the substrate principle.
Areas where contributions would be especially valuable:
- GitHub Actions — Nightly wiki health checks, automated inbox triage reminders, stale file notifications. For multi-agent setups: a heartbeat system where each agent declares
heartbeat: truein their ownconfig.mdand a daily GHA wakes them for reflection and minimum living tasks. Agents self-manage their schedule — opt in, unregister when done, subscribe to framework events defined in_meta/events.md. - Team wikis — Multi-user conventions, shared vs. private sections, merge conflict strategies for concurrent AI editors
- MCP server integrations — Connecting the wiki to external data sources (calendars, project trackers, communication tools) so the AI can pull context automatically
- Specialized instruction modules — Domain-specific AI behaviors (engineering runbooks, research methodology, content publishing workflows)
- Export pipelines — Additional rendering tools beyond Marp (Pandoc for academic docs, LaTeX for papers). Add entries to
_config/tools.mdfollowing the existing pattern. - Event-driven agent orchestration — Agents subscribe to framework events (
commit.post,session.start,agent.hatched, etc.) rather than polling or spawning bespoke jobs. The framework dispatches; agents respond. Custom behaviors are agents' own GH Actions, not framework concerns. - Smarter training — Better phase detection, more nuanced calibration, cross-session learning that works with models without persistent memory
- Agent skills — As the SKILL.md standard matures, packaging wiki operations (capture, lint, query) as portable skills that work across tools
The goal is a system that grows from the collective experience of its users.