agent-self-management

To record lessons learned for improving agent performance, identify the insight, classify it as a pattern, anti-pattern, gotcha, or optimization, determine the scope, write a concise note with context and evidence, and store it in the appropriate memory hierarchy.

The agent self-management memory hierarchy consists of user, session, and repository memories. It works by storing cross-workspace facts, current-task context, and codebase-specific knowledge respectively to persistently retain insights across different operational scopes.

Agents retrieve prior knowledge by checking memory at session start for relevant stored insights and cross-reference findings with known gotchas documented during previous post-resolution reviews to prevent repeated mistakes.

Yes, you can capture anti-patterns and document patterns within specific project or task contexts. The system supports structured memory capture with lesson typing and scope tagging applied across workspace, project, or task contexts.

A limitation of persistent memory storage for continuous agent improvement is that agents must actively check memory at session start and after post-resolution reviews; failure to retrieve or record lessons leads to repeated mistakes and stagnant behavior.