Looted/kibi

Kibi is a repo-local, per-git-branch, queryable knowledge base for software projects. It stores requirements, scenarios, tests, architecture decisions, and more as linked entities, ensuring end-to-end traceability between code and documentation.

Entity Taxonomy

Kibi intentionally supports eight core entity types, organized into two logical groups:

Common Authoring Entities

• req — Software requirements specifying functionality or constraints.
• scenario — BDD scenarios describing user behavior (Given/When/Then).
• test — Executable unit, integration, or e2e test cases.
• fact — Atomic domain facts and invariants. Supports a strict lane for contradiction-sensitive modeling and a context lane (observation, meta) for bugs and workarounds.

Supporting & System Entities

• adr — Architecture Decision Records documenting technical choices.
• flag — Runtime or config gates (feature flags, kill-switches).
• event — Domain or system events published/consumed by components.
• symbol — Abstract code symbols (functions, classes, modules).

Why Kibi

Kibi is designed to boost AI agents' memory during software development. It maintains a living, verifiable project memory that:

• Tracks context across branches — Every git branch gets its own KB snapshot, preserving context as you switch between features
• Enforces traceability — Links code symbols to requirements, preventing orphan features and technical debt
• Validates automatically — Rules catch missing requirements, dangling references, and consistency issues
• Agent-friendly — LLM assistants can query and update knowledge base via MCP without risking file corrupti