save-learning

You can persist user instructions by expressing a lasting preference, which the system refines into a structured learning with a type prefix and saves for future sessions. It classifies the instruction, requests explicit confirmation, and stages the update to CALIBER_LEARNINGS.md.

Remembering a convention works by proposing a clear action with a type prefix like [convention], [pattern], [anti-pattern], [preference], or [context]. The system requests explicit confirmation before saving the refined learning to ensure accuracy.

Yes, you can save learnings at either personal or project-wide scope levels. When adding a learning via the system command, you select the desired scope to ensure the instruction applies correctly to your future sessions.

To stop repeating setup instructions, tell the system to remember a specific preference or convention. It captures the instruction as a persistent learning, reducing repetitive setup and preventing mistakes in future sessions.

You should not use persistent learnings for temporary context or one-time instructions that should not apply to future sessions. They are designed exclusively for lasting instructions and preferences that need to persist across sessions.

Saving a learning stages updates to CALIBER_LEARNINGS.md for the next commit. It does not automatically commit the file, giving you control over when the staged changes are finalized in your version control system.