Agent Memory

Agent memory compounds knowledge across sessions instead of letting it evaporate. Because every session starts with an empty context window, Gemba Flow writes what matters to durable stores that the next session reads before picking up any work.

The problem

Without persistence, every agent session is groundhog day. It re-reads the same files, re-discovers the same patterns, and makes the same mistakes — at machine speed. The cost of relearning is paid every session.

The four layers

Memory entities

• CompletedTicket-{issue-number} — what shipped, where, and what changed.
• Pattern-{domain}-{short-name} — a reusable pattern found during work.
• Lesson-{domain}-{short-name} — a mistake and what we now do instead.
• Decision-{feature-name} — an architectural choice with rationale.

Agents query memory at task start (/work-ticket reads recent CompletedTickets), and append to memory at task end (/log-session writes LessonLearned and PatternDiscovered).