WW Analyze Skill

Deep analysis workflows for World Weaver memory systems, code quality, and architecture.

Purpose

This skill provides comprehensive analysis capabilities:

1. Code Analysis: Audit WW codebase for bugs, patterns, and improvements
2. Memory Analysis: Analyze memory contents, patterns, and health
3. Architecture Analysis: Evaluate system design and propose improvements
4. Performance Analysis: Profile and identify bottlenecks

When to Use

Invoke this skill when:

• User asks "analyze the memory system"
• User wants to understand memory patterns
• Code quality audit is needed
• Performance issues are suspected
• Architecture review is requested

Analysis Workflows

1. Bug Hunting Workflow

Orchestrate specialized bug-hunting agents:

# Run all bug hunters in sequence
paths=(
  "src/ww/learning/"
  "src/ww/memory/"
  "src/ww/storage/"
  "src/ww/mcp/"
  "src/ww/core/"
)

for path in "${paths[@]}"; do
  echo "Analyzing: $path"
done

Agent orchestration:

1. ww-bio-auditor - Check biological plausibility
2. ww-race-hunter - Find concurrency bugs
3. ww-leak-hunter - Detect memory leaks
4. ww-hinton-validator - Validate learning theory
5. ww-cache-analyzer - Check cache coherence
6. ww-trace-debugger - Debug eligibility traces

2. Memory Pattern Analysis

Analyze stored memories for patterns:

# Query memory statistics
mcp__ww-memory__memory_stats()

# Analyze episode distribution
mcp__ww-memory__recall_episodes(
  query="*",
  limit=1000,
  include_metadata=True
)

# Analyze entity graph
mcp__ww-memory__semantic_recall(
  query="*",
  include_connections=True
)

Output analysis:

• Episode count by outcome (success/failure/partial)
• Entity type distribution
• Relationship density
• Temporal patterns
• Importance distribution

3. Architecture Analysis

Evaluate system architecture:

# File structure analysis
find /home/aaron/ww/src -name "*.py" | wc -l

# Dependency analysis
grep -r "^from ww" /home/aaron/ww/src --include="*.py" | cut -d: -f2 | sort | uniq -c | sort -rn

# Test coverage check
cd /home/aaron/ww && pytest --cov=src/ww --cov-report=term-missing

Architecture metrics:

• Module coupling (import analysis)
• Test coverage by module
• Cyclomatic complexity
• Code duplication

4. Performance Analysis

Profile system performance:

import cProfile
import pstats

# Profile memory operations
profiler = cProfile.Profile()
profiler.enable()
# ... memory operations ...
profiler.disable()

stats = pstats.Stats(profiler)
stats.sort_stats('cumulative')
stats.print_stats(20)

Performance metrics:

• Query latency (p50, p95, p99)
• Memory usage over time
• CPU utilization
• I/O operations

Analysis Report Format

## WW Analysis Report

**Type**: {Bug Hunt | Memory Pattern | Architecture | Performance}
**Date**: {timestamp}
**Scope**: {paths analyzed}

### Summary
{High-level findings}

### Metrics
| Metric | Value | Status |
|--------|-------|--------|
| Files analyzed | N | - |
| Issues found | N | {OK/WARNING/CRITICAL} |
| Test coverage | N% | {OK if >80%} |

### Findings

#### Critical (P0)
{List of critical issues}

#### High (P1)
{List of high priority issues}

#### Medium (P2)
{List of medium priority issues}

### Recommendations
1. {Priority action items}

### Visualizations
{Embedded diagrams or links to generated visualizations}

Integration with Agents

This skill orchestrates bug-hunting agents:

/ww-analyze bugs src/ww/learning/
  → Spawns: ww-bio-auditor, ww-hinton-validator, ww-trace-debugger

/ww-analyze concurrency src/ww/mcp/
  → Spawns: ww-race-hunter, ww-leak-hunter, ww-cache-analyzer

/ww-analyze full src/ww/
  → Spawns: All 6 agents in parallel

MCP Extensions

Proposed MCP endpoints for analysis:

mcp__ww-memory__analyze_patterns    - Analyze memory patterns
mcp__ww-memory__analyze_health      - Check system health
mcp__ww-memory__analyze_performance - Profile operations
mcp__ww-memory__generate_report     - Create analysis report

Quality Checklist

Before completing analysis:

• All target paths scanned
• All agents completed successfully
• Findings categorized by severity
• Recommendations are actionable
• Report saved to /home/aaron/mem/

Error Handling

If analysis fails:

1. Log partial results
2. Identify failing component
3. Continue with remaining analyses
4. Report incomplete status