Roadmap Position: Step 5 of 6 — historical-learning strengthening layer
Depends on: #123, #124, #125, #126, plus cron infrastructure
Enables: automatic execution of the retrospective learning pipeline

Parent Epic: #138 (Cron Infrastructure)
Related Epic: #116 (Historical Learning — this issue is the bridge)

Problem

MemoryConsolidator exists but only runs when manually called. PRD requires automatic triggers: after every N sessions, when memory exceeds size threshold, and during daemon idle.

Solution

Integration with CronScheduler

Register consolidation jobs during daemon startup:

// In AceClawDaemon.start()
cronScheduler.register(new CronJob(
    "memory-consolidation",
    "*/6h",  // every 6 hours
    () -> {
        memoryConsolidator.consolidate(autoMemoryStore, projectPath, archiveDir);
        sessionAnalyzer.analyzeRecentSessions(transcriptStore, indexer);
        crossSessionMiner.mine(index, memory, 20);
        trendDetector.detect(index, 10);
    }
));

Triggers

1. Time-based: every 6 hours via CronScheduler
2. Session-count: after every 10 session closes
3. Size-based: when any memory JSONL exceeds 50KB
4. Idle-based: when daemon has no active sessions for 5 minutes

Pipeline Order

1. MemoryConsolidator (dedup + merge + prune)
2. SessionAnalyzer (analyze un-analyzed sessions)
3. HistoricalLogIndex rebuild (if stale)
4. CrossSessionPatternMiner
5. TrendDetector
6. Feed results to AutoMemoryStore + SkillProposalEngine

Dependencies

• Depends on feat(cron-p0): add precise schedule kinds (at/everyMs) with deterministic nextFire #112 (precise scheduling) for everyMs sub-minute triggers
• Depends on feat(cron-p1): bounded parallel execution with queueing and non-reentrant job safety #113 (parallel execution) if consolidation runs alongside other cron jobs
• Outputs consumed by feat(learning): SessionAnalyzer for full session retrospective analysis #123 SessionAnalyzer, feat(learning): CrossSessionPatternMiner — mine recurring patterns across sessions #125 CrossSessionPatternMiner, feat(learning): TrendDetector — detect rising/falling metrics over time #126 TrendDetector

Acceptance Criteria

• Consolidation runs automatically on schedule (no manual trigger needed)
• Session-count trigger works (every 10 sessions)
• Size-based trigger works (50KB threshold)
• Full pipeline executes in order on virtual thread