memory-management

Agent memory organization separates recent activity from longer-term patterns using a tri-tier system. This prevents context loss by moving information across hot, warm, and cold layers as task history grows over time.

State consolidation for long-running workflows uses weekly cleanup to move records from hot to warm to cold memory layers. This keeps agent memory concise, searchable, and useful as operational history expands.

Recording agent learnings after finishing work uses hot memory updates to log recent events, active tasks, and fresh outcomes. This structured memory update captures results immediately after task completion.

No external dependencies are needed to manage agent memory. The skill uses safe script-based introspection with YAML to handle memory updates and promotion rules between tiers independently.

Archiving agent memory to cold storage happens when recent events and patterns grow stale. The tri-tier system promotes information from hot to warm to cold based on retention rules for long-running workflows.

Yes, this memory management approach is designed for long-running agent workflows. It supports operational cleanup and pattern consolidation to ensure memory stays organized and current during continuous task execution.