Knowledge Base

Protocol-driven documentation and knowledge management for multi-agent startup development.

Quick Start

For Humans

1. Open docs/ as an Obsidian vault
2. Read CLAUDE.md for project conventions
3. Browse docs/architecture/ for system context

For AI Agents

1. Read CLAUDE.md — this is your operating protocol
2. Read docs/meta/OWNERSHIP.md — know your boundaries
3. Work on your assigned branch, submit PRs with intent declarations

Structure

docs/                  Obsidian vault — the knowledge base
├── architecture/      System design docs
├── api/               API reference
├── adr/               Architecture Decision Records
├── runbooks/          Operational playbooks
└── meta/              Ownership, drift log, conventions
scripts/ci/            CI consistency checks
scripts/migrate/       Platform migration tools
backup/                Backup scripts
CLAUDE.md              Agent protocol (read-only for agents)

Key Concepts

• Protocol-first: Agent behavior is defined by versioned Markdown files, not code
• Falsifiable trust: We detect agent deviations through auditable logs, not self-explanation
• Failure as data: Every protocol violation is recorded and drives protocol improvement
• Partitioned ownership: Each agent/human owns specific doc directories to reduce conflicts

CI Checks

Every PR touching docs/ or src/ triggers:

• Orphan backlink detection
• Front matter completeness
• Code-doc accompaniment check
• Ownership boundary warnings

Backup

3-2-1 strategy: GitHub (primary) + Backblaze B2 (cold) + local clones. See backup/rclone-backup.sh.

References

• ADR-001: Knowledge Base Architecture
• PRD
• EDD