Continuous Learning System

You can persist agent debugging context across sessions by capturing project-specific learnings into JSONL files. The system stores what worked, what broke, and why under a project-scoped directory so future runs reload relevant context instead of repeating prior work.

Loading prior learnings at session start restores the last session state and injects high-confidence items (0.8+) into agent context. This speeds up reasoning and reduces repeated mistakes by immediately providing historical architecture, debugging, and tool usage knowledge.

The best way to record root causes and pitfalls is structured learning capture at session boundaries. The system logs discoveries across patterns, pitfalls, architecture decisions, and tool behaviors, then applies confidence scoring to accumulate reliable guidance over time.

Yes, you can use confidence scoring to manage accumulated agent learnings. The system applies confidence-based accumulation to project-scoped JSONL storage, ensuring guidance improves with ongoing usage and only high-relevance context is injected into future agent sessions.

Project learnings are stored as JSONL files under the directory path ~/.buildplayplay/projects/{project-slug}/. This project-scoped storage structure ensures session-end persistence separates learnings per project using stored git state.

You do not need to manually save session outcomes. The system automatically handles session-end persistence to project-scoped storage, while session-start learning loading automatically retrieves and injects relevant prior learnings into your agent workflow.