prune-learnings

Prune Learnings Command

Manually invokes the pruning script to clean up old learnings and rotate log files.

Purpose

Over time, the learnings system accumulates:

• Old session directories
• Large log files
• Archived pending files

This command runs the pruning script to clean up old data according to retention policy.

Retention Configuration

Edit .learnings/retention.yaml to customize:

# Maximum number of runs to keep in runs.log
max_runs: 100

# Maximum number of session directories to keep
max_sessions: 50

# Maximum lines per log file before rotation
max_log_lines: 10000

# Maximum age (days) for archived pending files
max_archive_age_days: 30

# Lock staleness threshold (hours) before force-pruning
lock_stale_hours: 4

# Protected run window (minutes) - recent runs won't be pruned
protected_window_minutes: 30

What Gets Pruned

1. Session directories: Oldest sessions beyond max_sessions limit
2. Log files: Rotated when exceeding max_log_lines
   • runs.log → runs.log.1 → deleted
   • outcomes.log → outcomes.log.1 → deleted
   • acknowledgments.log → acknowledgments.log.1 → deleted
3. Archived pending files: Older than max_archive_age_days
4. Stale lock files: Lock files older than lock_stale_hours

Safety Mechanisms

• Protected runs: Current run and runs active within protected_window_minutes are never pruned
• Lock checking: Won't prune if .learnings/.lock exists (unless stale)
• Atomic operations: Uses atomic file replacement to prevent corruption

Usage

# Run pruning script directly
./plugins/self-learning/scripts/prune-learnings.sh

# Or via ClosedLoop orchestrator command
# This is automatically run after each completed run

Automatic Pruning

Pruning runs automatically:

1. After each run completes (in background)
2. Current run is always protected
3. Failures are logged but don't block the run

Manual Pruning

Use manual pruning when:

• Disk space is low
• You want to clean up before sharing repository
• You've changed retention settings and want immediate effect