MyKnowledge

Problem

Multi-step development, research, and data-cleanup work often scatters requirements, status, conclusions, and materials across chat history, temporary files, and single sessions. When a project resumes, the AI may not know what was already completed, and manually maintained docs can fall behind.

How It Works

MyKnowledge treats the knowledge base as a manageable file structure rather than only prompt conventions:
- Initialization: on first load, it checks skill-state.yaml; if missing, it runs a short guided setup and saves configuration.
- Knowledge base creation: supports global and project structures; users can create one explicitly, and complex tasks can trigger automatic creation with a notice.
- Requirement management: tracks status transitions across Created, In Progress, Review, Done, and Cancelled, while maintaining a PROJECT-STATUS.md snapshot.
- Migration and recovery: supports export/import and session recovery for sharing, backup, and context restoration.
- Platform adaptation: OpenClaw uses Hooks on message receipt to detect complexity, while CodeBuddy/WorkBuddy and Claude rely on intent recognition to decide whether a knowledge base is needed.

Boundaries and Notes

The default path is ~/.myknowledge/, and the skill only accesses that authorized directory rather than other system settings. Automatic detection depends on keywords and task features, so only sufficiently complex tasks trigger it. OpenClaw Hook mode runs in the background and notifies after the action; it is not fully invisible. Claude Hook support depends on the specific environment.