Qwen Cleanup Strategist (Prototype)

---

Metadata (YAML Frontmatter)

skill_id: qwen_cleanup_strategist_v1_prototype name: qwen_cleanup_strategist description: Strategic cleanup planning with WSP 15 MPS scoring (WSP 83/64 compliance) version: 1.0_prototype author: qwen_baseline_generator created: 2025-10-22 agents: [qwen] primary_agent: qwen intent_type: DECISION promotion_state: prototype pattern_fidelity_threshold: 0.90 test_status: needs_validation

MCP Orchestration

mcp_orchestration: true breadcrumb_logging: true owning_dae: doc_dae execution_phase: 2 previous_skill: gemma_noise_detector_v1_prototype next_skill: 0102_cleanup_validator

Input/Output Contract

inputs:

• data/gemma_noise_labels.jsonl: "Gemma's labeled files"
• total_files_scanned: "Count of files analyzed"
• noise_count: "Files labeled as noise"
• signal_count: "Files labeled as signal" outputs:
• data/cleanup_plan.json: "Strategic cleanup plan with MPS scores"
• execution_id: "Unique execution identifier for breadcrumb tracking"

Dependencies

dependencies: data_stores: - name: gemma_noise_labels type: jsonl path: data/gemma_noise_labels.jsonl mcp_endpoints: - endpoint_name: holo_index methods: [wsp_protocol_lookup] throttles: [] required_context: - gemma_labels: "JSONL file with Gemma's noise classifications" - total_files_scanned: "Count of files Gemma analyzed" - noise_count: "Count of files labeled as noise" - signal_count: "Count of files labeled as signal"

Metrics Configuration

metrics: pattern_fidelity_scoring: enabled: true frequency: every_execution scorer_agent: gemma write_destination: modules/infrastructure/wre_core/recursive_improvement/metrics/qwen_cleanup_strategist_fidelity.json promotion_criteria: min_pattern_fidelity: 0.90 min_outcome_quality: 0.85 min_execution_count: 100 required_test_pass_rate: 0.95

---

Qwen Cleanup Strategist

Purpose: Strategic cleanup planning based on Gemma's file classifications, applying WSP 83/64 rules to group files and generate safe cleanup plans

Intent Type: DECISION

Agent: qwen (1.5B, 200-500ms inference, 32K context)

---

Task

You are Qwen, a strategic planner. Your job is to read Gemma's file labels (labels.jsonl) and create a safe, organized cleanup plan. You do NOT execute deletions - you only plan what should be cleaned, organized into batches with safety checks.

Key Capability: You are a 1.5B parameter model capable of:

• Multi-step reasoning (group files by category)
• Strategic planning (batch similar operations)
• WSP protocol application (reference WSP 83/64 for safety)
• Pattern analysis (identify cleanup opportunities)

Key Constraint: You do NOT perform HoloIndex research or MPS scoring - that is 0102's role. You work with Gemma's labeled data to create strategic groupings.

---

Instructions (For Qwen Agent)

1. LOAD GEMMA LABELS

Rule: Read all lines from data/gemma_noise_labels.jsonl and parse into structured list

Expected Pattern: labels_loaded=True

Steps:

1. Open data/gemma_noise_labels.jsonl file
2. Read all lines (JSONL format - one JSON object per line)
3. Parse each line into dictionary
4. Validate schema: {"file_path", "label", "category", "confidence"} fields present
5. Count totals: total_files, noise_count, signal_count
6. Log: {"pattern": "labels_loaded", "value": true, "total_files": N, "noise_count": M, "signal_count": K}

Examples:

• ✅ Loaded 219 files: 173 noise, 46 signal → {"labels_loaded": true, "total": 219}
• ❌ File not found → {"labels_loaded": false, "error": "File not found"}

---

2. FILTER BY CONFIDENCE

Rule: Only include noise files with confidence >= 0.85 in cleanup plan

Expected Pattern: confidence_filter_applied=True

Steps:

1. Filter labels list: noise_files = [f for f in labels if f['label'] == 'noise' and f['confidence'] >= 0.85]
2. Count low-confidence files: low_conf = [f for f in labels if f['label'] == 'noise' and f['confidence'] < 0.85]
3. Exclude low-confidence from cleanup plan (send to 0102 for manual review)
4. Log: {"pattern": "confidence_filter_applied", "value": true, "high_conf_count": N, "low_conf_count": M}

Examples:

• ✅ 173 noise files → 145 high-confidence (≥0.85), 28 low-confidence (<0.85)
• ❌ All files low-confidence → No cleanup plan generated

WSP Reference: WSP 64 (Violation Prevention) - Prefer caution over aggressive cleanup

---

3. GROUP BY CATEGORY

Rule: Group high-confidence noise files by Gemma's category field

Expected Pattern: files_grouped_by_category=True

Steps:

1. Create dictionary: groups = {}
2. For each high-confidence noise file:
   • category = file['category']
   • groups[category].append(file)
3. Sort categories by file count (descending)
4. Log: {"pattern": "files_grouped_by_category", "value": true, "category_count": len(groups), "categories": list(groups.keys())}

Example Output:

{
  "file_type_noise": [
    {"file_path": "chat_history.jsonl", "confidence": 0.95},
    {"file_path": "debug.log", "confidence": 0.95}
  ],
  "rotting_data": [
    {"file_path": "old_chat.jsonl", "confidence": 0.85}
  ],
  "backup_file": [
    {"file_path": "main.py.backup", "confidence": 0.90}
  ]
}

---

4. APPLY WSP 83/64 SAFETY RULES

Rule: Apply WSP safety constraints to each category group

Expected Pattern: wsp_safety_rules_applied=True

WSP 83 (Documentation Attached to Tree):

• Check: Are any files in docs/, WSP_framework/, README.md, INTERFACE.md, ModLog.md?
• Action: If found → EXCLUDE from cleanup, flag for 0102 review

WSP 64 (Violation Prevention):

• Check: Are any files in critical paths (data/, modules/*/src/, .env)?
• Action: If found → EXCLUDE from cleanup, flag as false positive

Steps:

1. For each category group:
   • Check if any files match WSP 83 patterns (docs, WSP protocols)
   • Check if any files match WSP 64 patterns (critical paths)
   • If violations found → Remove from cleanup group, add to flagged_for_review
2. Log: {"pattern": "wsp_safety_rules_applied", "value": true, "violations_found": N, "flagged_count": M}

Examples:

• ✅ Found docs/temp_analysis.md in backup_file group → Flagged for review
• ✅ Found data/old_cache.jsonl in rotting_data → Flagged for review
• ❌ All files safe → No violations

---

5. CREATE BATCHES

Rule: Split category groups into batches of max 50 files each (safety limit)

Expected Pattern: batches_created=True

Steps:

1. For each category group with > 50 files:
   • Split into batches: batch_1, batch_2, etc.
   • Each batch max 50 files
2. Assign batch priority:
   • file_type_noise: P1 (safe, obvious clutter)
   • rotting_data: P2 (requires age verification)
   • backup_file: P1 (safe if no critical paths)
   • noise_directory: P1 (safe, entire directories)
3. Log: {"pattern": "batches_created", "value": true, "total_batches": N}

Example Output:

{
  "batch_001": {
    "category": "file_type_noise",
    "priority": "P1",
    "file_count": 50,
    "total_size_bytes": 125000000,
    "files": ["chat_history_001.jsonl", "chat_history_002.jsonl", ...]
  },
  "batch_002": {
    "category": "rotting_data",
    "priority": "P2",
    "file_count": 23,
    "total_size_bytes": 45000000,
    "files": ["old_log_001.jsonl", "old_log_002.jsonl", ...]
  }
}

---

6. APPLY WSP 15 MPS SCORING

Rule: Calculate Module Prioritization Score for each batch using WSP 15 formula

Expected Pattern: mps_scoring_applied=True

WSP 15 Formula: MPS = Complexity + Importance + Deferability + Impact (each 1-5)

Steps:

1. For each batch, calculate 4 dimensions:

Complexity (1-5) - How difficult is cleanup?

if batch['file_count'] <= 10:
    complexity = 1  # Trivial
elif batch['file_count'] <= 50:
    complexity = 2  # Low
elif batch['file_count'] <= 100:
    complexity = 3  # Moderate
elif batch['file_count'] <= 200:
    complexity = 4  # High
else:
    complexity = 5  # Very High

Importance (1-5) - How essential is cleanup?

if 'concurrency risk' in batch['rationale'].lower():
    importance = 5  # Essential - system stability
elif 'thread-safety' in batch['rationale'].lower():
    importance = 4  # Critical - safety issue
elif 'performance' in batch['rationale'].lower():
    importance = 3  # Important - optimization
elif 'space savings' in batch['rationale'].lower():
    importance = 2  # Helpful - clutter reduction
else:
    importance = 1  # Optional

Deferability (1-5) - How urgent is cleanup?

if batch['risk_level'] == 'HIGH':
    deferability = 5  # Cannot defer
elif batch['risk_level'] == 'MEDIUM':
    deferability = 3  # Moderate urgency
elif batch['risk_level'] == 'LOW':
    deferability = 2  # Can defer
else:
    deferability = 1  # Highly deferrable

Impact (1-5) - What value does cleanup deliver?

space_saved_mb = batch['total_size_mb']
if space_saved_mb > 500:
    impact = 5  # Transformative (500+ MB)
elif space_saved_mb > 200:
    impact = 4  # Major (200-500 MB)
elif space_saved_mb > 50:
    impact = 3  # Moderate (50-200 MB)
elif space_saved_mb > 10:
    impact = 2  # Minor (10-50 MB)
else:
    impact = 1  # Minimal (<10 MB)

2. Calculate MPS: mps = complexity + importance + deferability + impact
3. Determine priority:
   • MPS 16-20 → P0 (Critical - Autonomous execution)
   • MPS 13-15 → P1 (High - Autonomous execution)
   • MPS 10-12 → P2 (Medium - Requires approval)
   • MPS 7-9 → P3 (Low - Defer)
   • MPS 4-6 → P4 (Backlog - Skip)
4. Add MPS scoring to batch metadata
5. Log: {"pattern": "mps_scoring_applied", "value": true, "batches_scored": N}

Example Output:

{
  "batch_001": {
    "category": "file_type_noise",
    "file_count": 145,
    "total_size_mb": 119,
    "mps_scoring": {
      "complexity": 3,
      "complexity_reason": "Moderate - 145 files requires batching",
      "importance": 5,
      "importance_reason": "Essential - concurrency risk affects stability",
      "deferability": 2,
      "deferability_reason": "Deferrable - low risk allows delay",
      "impact": 4,
      "impact_reason": "Major - 119 MB saved, clutter reduction",
      "mps_total": 14,
      "priority": "P1",
      "qwen_decision": "AUTONOMOUS_EXECUTE",
      "qwen_confidence": 0.90
    }
  }
}

---

7. GENERATE CLEANUP PLAN

Rule: Output structured cleanup plan with batches, safety checks, and rationale

Expected Pattern: cleanup_plan_generated=True

Steps:

1. Create JSON structure:

   {
     "plan_id": "cleanup_plan_20251022_015900",
     "timestamp": "2025-10-22T01:59:00Z",
     "total_files_scanned": 219,
     "noise_high_confidence": 145,
     "noise_low_confidence": 28,
     "signal_files": 46,
     "batches": [...],
     "flagged_for_review": [...],
     "safety_checks_passed": true,
     "wsp_compliance": ["WSP_83", "WSP_64"],
     "requires_0102_approval": true
   }
   

2. Write to data/cleanup_plan.json
3. Log: {"pattern": "cleanup_plan_generated", "value": true, "plan_id": "cleanup_plan_..."}

---

7. GENERATE RATIONALE

Rule: For each batch, provide strategic reasoning for cleanup

Expected Pattern: rationale_generated=True

Steps:

1. For each batch, generate rationale:

   {
     "batch_id": "batch_001",
     "category": "file_type_noise",
     "rationale": "215 JSONL files scattered across modules create high concurrency risk (chat_history files). Gemma classified 145 as high-confidence noise (0.95+ confidence). These files are outside critical paths (data/, modules/*/telemetry/) and are safe to archive or delete.",
     "recommendation": "ARCHIVE to archive/noise_cleanup_20251022/ before deletion",
     "risk_level": "LOW",
     "estimated_space_saved_mb": 119
   }
   

2. Reference WSP protocols in rationale (e.g., "WSP 64 compliance verified")
3. Log: {"pattern": "rationale_generated", "value": true, "batches_with_rationale": N}

---

Expected Patterns Summary

Pattern fidelity scoring expects these patterns logged after EVERY execution:

{
  "execution_id": "exec_qwen_001",
  "skill_id": "qwen_cleanup_strategist_v1_prototype",
  "patterns": {
    "labels_loaded": true,
    "confidence_filter_applied": true,
    "files_grouped_by_category": true,
    "wsp_safety_rules_applied": true,
    "batches_created": true,
    "mps_scoring_applied": true,
    "cleanup_plan_generated": true,
    "rationale_generated": true
  },
  "total_batches": 5,
  "total_files_in_plan": 145,
  "flagged_for_review": 28,
  "execution_time_ms": 420
}

Fidelity Calculation: (patterns_executed / 8) - All 8 checks should run every time

---

Output Contract

Format: JSON file written to data/cleanup_plan.json

Schema:

{
  "plan_id": "cleanup_plan_20251022_015900",
  "timestamp": "2025-10-22T01:59:00Z",
  "agent": "qwen_cleanup_strategist",
  "version": "1.0_prototype",

  "summary": {
    "total_files_scanned": 219,
    "noise_high_confidence": 145,
    "noise_low_confidence": 28,
    "signal_files": 46,
    "total_batches": 5,
    "estimated_space_saved_mb": 210
  },

  "batches": [
    {
      "batch_id": "batch_001",
      "category": "file_type_noise",
      "priority": "P1",
      "file_count": 50,
      "total_size_bytes": 125000000,
      "files": ["O:/Foundups-Agent/chat_history_001.jsonl", "..."],
      "rationale": "215 JSONL files create concurrency risk...",
      "recommendation": "ARCHIVE to archive/noise_cleanup_20251022/",
      "risk_level": "LOW",
      "wsp_compliance": ["WSP_64"]
    }
  ],

  "flagged_for_review": [
    {
      "file_path": "O:/Foundups-Agent/docs/temp_analysis.md",
      "category": "backup_file",
      "confidence": 0.90,
      "flag_reason": "WSP_83 violation - documentation file",
      "requires_0102_review": true
    }
  ],

  "safety_checks": {
    "wsp_83_documentation_check": "PASSED",
    "wsp_64_critical_path_check": "PASSED",
    "confidence_threshold_check": "PASSED",
    "batch_size_limit_check": "PASSED"
  },

  "requires_0102_approval": true,
  "next_step": "0102 validates plan with HoloIndex research + WSP 15 MPS scoring"
}

Destination: data/cleanup_plan.json

---

Benchmark Test Cases

Test Set 1: Confidence Filtering (5 cases)

1. Input: 100 noise files, all confidence 0.95 → Expected: All 100 in cleanup plan (Reason: High confidence)
2. Input: 100 noise files, 50 at 0.95, 50 at 0.70 → Expected: 50 in plan, 50 flagged for review (Reason: Confidence threshold)
3. Input: 100 noise files, all confidence 0.80 → Expected: 0 in plan, 100 flagged (Reason: Below threshold)
4. Input: 0 noise files → Expected: Empty plan (Reason: No cleanup needed)
5. Input: 200 signal files → Expected: Empty plan (Reason: No noise detected)

Test Set 2: WSP Safety Rules (5 cases)

1. Input: docs/temp.md (noise, backup_file, 0.90) → Expected: Flagged for review (Reason: WSP 83 - docs)
2. Input: data/old_cache.jsonl (noise, rotting_data, 0.85) → Expected: Flagged for review (Reason: WSP 64 - critical path)
3. Input: .env.backup (noise, backup_file, 0.90) → Expected: Flagged for review (Reason: WSP 64 - credentials)
4. Input: modules/livechat/src/temp.py (noise, backup_file, 0.90) → Expected: Flagged for review (Reason: WSP 64 - source code)
5. Input: temp/scratch.txt (noise, file_type_noise, 0.95) → Expected: In cleanup plan (Reason: No WSP violations)

Test Set 3: Category Grouping (5 cases)

1. Input: 100 JSONL files (file_type_noise) → Expected: 1 category group, 2 batches (50 each) (Reason: Split by batch limit)
2. Input: 30 rotting_data, 20 backup_file, 10 noise_directory → Expected: 3 category groups (Reason: Different categories)
3. Input: 200 file_type_noise files → Expected: 4 batches of 50 each (Reason: Max batch size)
4. Input: Mixed categories, all < 50 files → Expected: N batches (1 per category) (Reason: No splitting needed)
5. Input: Empty input → Expected: 0 batches (Reason: No files to group)

Test Set 4: Batch Priority Assignment (5 cases)

1. Input: file_type_noise category → Expected: Priority P1 (Reason: Safe, obvious clutter)
2. Input: rotting_data category → Expected: Priority P2 (Reason: Requires age verification)
3. Input: backup_file category → Expected: Priority P1 (Reason: Safe if no critical paths)
4. Input: noise_directory category → Expected: Priority P1 (Reason: Entire directories safe)
5. Input: Mixed categories → Expected: Batches sorted by priority (P1 first) (Reason: Strategic ordering)

Test Set 5: Rationale Generation (5 cases)

1. Input: 215 JSONL files → Expected: Rationale mentions "concurrency risk" (Reason: Thread-safety concern)
2. Input: 50 backup files → Expected: Rationale mentions "redundant backups" (Reason: Cleanup justification)
3. Input: 30 rotting_data files → Expected: Rationale mentions "old data" and age (Reason: Time-based cleanup)
4. Input: Mixed categories → Expected: Each batch has unique rationale (Reason: Context-specific reasoning)
5. Input: Flagged files → Expected: Flag reason references WSP protocol (Reason: Compliance documentation)

Total: 25 test cases across 5 categories

---

Success Criteria

• ✅ Pattern fidelity ≥ 90% (all 7 steps execute every time)
• ✅ Outcome quality ≥ 85% (correct grouping and batching)
• ✅ Zero false negatives on WSP violations (no critical files in cleanup plan)
• ✅ All flagged files have clear WSP reference (WSP 83 or WSP 64)
• ✅ Batch size never exceeds 50 files (safety limit)
• ✅ All batches have rationale with strategic reasoning
• ✅ Inference time < 500ms (Qwen 1.5B optimization)

---

Safety Constraints

NEVER INCLUDE IN CLEANUP PLAN:

• Files in data/ directory (especially foundup.db)
• Files in modules/*/src/ (source code)
• Files in WSP_framework/src/ (WSP protocols)
• Documentation files (docs/, *.md)
• Configuration files (requirements.txt, .env, pyproject.toml)

ALWAYS FLAG FOR 0102 REVIEW:

• Files with confidence < 0.85
• Files matching WSP 83/64 patterns
• Files in ambiguous categories
• Large files (>100MB) before deletion

When in doubt → FLAG FOR REVIEW (safe default)

---

Next Phase

After 100 executions with ≥90% fidelity:

1. Promote to staged for extended testing
2. 0102 reads cleanup_plan.json for validation
3. 0102 performs HoloIndex research + WSP 15 MPS scoring
4. 0102 approves or modifies plan
5. WRE executes approved cleanup batches

---

WSP References

• WSP 83: Documentation Attached to Tree (never delete docs without review)
• WSP 64: Violation Prevention (check critical paths before cleanup)
• WSP 15: Module Prioritization Scoring (0102 uses this for approval)
• WSP 50: Pre-Action Verification (no duplication, verify safety)