What problem does it solve?
Removes the manual, ad-hoc process of reviewing agent sessions and maintaining long-term memory by automating nightly and hourly review loops that extract learnings and update memory files so the agent improves over time.
Core Features & Use Cases
- Automated Nightly Review: Scan all sessions from the last 24 hours, extract key learnings, and update long-term and daily memory files.
- Hourly Snapshots: Create frequent memory snapshots to capture incremental progress and decisions.
- Extracted Artifacts: Identify patterns, gotchas, user preferences, key decisions, and open items for follow-up.
- Integrations: Designed to run via cron, Clawdbot, or launchd and commit memory updates to git for versioned agent instructions.
- Use Case: Run a nightly compound job to consolidate a day's sessions into MEMORY.md and memory/YYYY-MM-DD.md so tomorrow's agent benefits from yesterday's insights.
Quick Start
Run the compound-engineering review command to scan the last 24 hours, extract key learnings, and update MEMORY.md and today's memory file.