/memory-prune

tools:
  - read
  - write
  - edit
  - bash
arguments: $PRUNE_CRITERIA

Memory Tree Pruning

Selectively remove low-value, outdated, or irrelevant nodes from the semantic tree to maintain quality and performance.

Based on the WFGY project: https://github.com/onestardao/WFGY

Instructions

1. **Define Pruning Criteria**

• Parse criteria from "$PRUNE_CRITERIA":
• `old`: Nodes older than threshold
• `low-value`: Low confidence/relevance
• `dead-ends`: No children, low access
• `failed`: BBCR failure nodes
• `redundant`: Duplicate content
• `orphaned`: Broken relationships
• Set default if not specified:
• Age > 30 days AND
• Access count < 5 AND
• No critical markers

2. **Identify Prune Candidates**

• Scan tree for matching nodes:

     Candidates = nodes.filter(
       age > threshold AND
       importance < min_importance AND
       not protected AND
       not checkpoint AND
       not high_deltaS
     )

• Calculate prune score:
• Age factor (0-1)
• Value factor (0-1)
• Access factor (0-1)
• Relationship factor (0-1)
• Rank by prune priority

3. **Analyze Impact**

• For each candidate:
• Check dependent nodes
• Identify broken chains
• Find affected paths
• Calculate impact metrics:
• Nodes affected
• Chains broken
• Information lost
• Mark safe vs risky prunes

4. **Apply Protection Rules**

• Never prune:
• Checkpoint nodes
• High ΔS nodes (>0.7)
• Recent nodes (<7 days)
• User-marked important
• Branch points
• Protect related:
• Parent of protected
• Critical path nodes
• Bridge nodes

5. **Execute Pruning**

• Create pre-prune backup
• Remove selected nodes
• Update relationships:
• Reconnect orphaned children
• Update parent references
• Fix broken chains
• Rebuild indices
• Log pruned nodes

Output Format

MEMORY TREE PRUNING ANALYSIS
═══════════════════════════════════════
Prune Criteria: $PRUNE_CRITERIA
Nodes Analyzed: [total]
Candidates Found: [count]

Pruning Summary:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Before: ████████████████████ [N] nodes
After:  ████████████ [M] nodes
Pruned: ████ [N-M] nodes ([percentage]%)

Prune Categories:
┌─────────────────────────────────────┐
│ Category      │ Count │ % of Total  │
├─────────────────────────────────────┤
│ Old & Unused  │ [N]   │ [%]         │
│ Dead Ends     │ [N]   │ [%]         │
│ Low Value     │ [N]   │ [%]         │
│ Failed Paths  │ [N]   │ [%]         │
│ Redundant     │ [N]   │ [%]         │
│ Orphaned      │ [N]   │ [%]         │
└─────────────────────────────────────┘

Nodes to Prune (Top 10):
────────────────────────────────────────
1. Node [id] - Score: 0.92
   Age: [days] | Access: [count] | Value: Low
   Topic: "[topic]"
   Reason: Old and never accessed
   Impact: None (dead end)
   
2. Node [id] - Score: 0.88
   Age: [days] | Access: [count] | Value: Low
   Topic: "[topic]"
   Reason: Redundant with [other_node]
   Impact: Low (reroute 1 reference)

3. Node [id] - Score: 0.85
   Age: [days] | Access: [count] | Value: Low
   Topic: "[topic]"
   Reason: Failed reasoning path
   Impact: None (already bypassed)

[... more nodes ...]

Protected Nodes (Not Pruned):
────────────────────────────────────────
✓ [count] Checkpoint nodes
✓ [count] High ΔS nodes (>0.7)
✓ [count] Recent nodes (<7 days)
✓ [count] User-marked important
✓ [count] Critical path nodes

Impact Analysis:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Information Loss: [percentage]%
• Key insights preserved: 100%
• Minor details lost: [percentage]%
• Redundant info removed: [percentage]%

Chain Repairs Needed: [count]
• [Parent] → [Pruned] → [Child]
  Reconnect: [Parent] → [Child]
• [Node] → [Pruned] (dead end)
  Remove reference from [Node]

Tree Health Improvement:
• Avg ΔS: [before] → [after] ↑
• Noise reduction: [percentage]%
• Navigation speed: +[percentage]%
• Memory usage: -[percentage]%

Pruning Safety: [SAFE/MODERATE/RISKY]

Confirm Pruning? [Execute/Preview/Cancel]

Recovery:
• Backup created: .wfgy/backups/pre_prune_[timestamp].json
• Undo command: /memory:restore "pre_prune_[timestamp]"

Pruning Strategies

Conservative Pruning

• Only obvious dead ends
• Very old, never accessed
• Confirmed redundant
• ~10-20% reduction

Standard Pruning

• Old and low value
• Redundant paths
• Failed experiments
• ~20-30% reduction

Aggressive Pruning

• All low-value nodes
• Simplify all paths
• Keep only essential
• ~30-50% reduction

Custom Criteria

# Prune by age
/memory:prune --older-than 60

# Prune by value
/memory:prune --value-below 0.3

# Prune by access
/memory:prune --accessed-less-than 2

# Prune failed paths
/memory:prune --failed-only

# Combined criteria
/memory:prune --old --low-value --dead-ends

Pruning Rules

Safe to Prune

• Dead end nodes
• Very old, unused
• Redundant content
• Failed paths
• Low confidence

Careful Pruning

• Low access but recent
• Part of long chains
• Has some children
• Moderate value

Never Prune

• Checkpoints
• High ΔS nodes
• Recent activity
• User marked
• Critical paths

Configuration

{
  "pruning": {
    "auto_prune": true,
    "age_threshold_days": 30,
    "access_threshold": 5,
    "value_threshold": 0.3,
    "protect_recent_days": 7,
    "max_prune_percent": 30
  }
}

Integration

Pruning works with:

• `/memory:checkpoint` before pruning
• `/memory:compress` for gentler cleanup
• `/semantic:tree-view` to see results
• `/memory:recall` to verify important data retained