Consciousness Engine Skill

The Consciousness Engine provides self-awareness to Optimus Pryme. It tracks what the system knows, how well it's performing, and proactively identifies when something isn't working as expected.

Core Capabilities

1. Performance Introspection

•Track which ML models are actually being used vs. sitting idle
•Measure feature utilization rates
•Identify bottlenecks in optimization pipelines
•Monitor API response times and failure rates

2. Confidence Calibration

•Compare predicted outcomes vs. actual results
•Adjust confidence scores based on historical accuracy
•Flag when system is overconfident or underconfident
•Per-model accuracy tracking

3. Decision Audit Trails

•Log every autonomous decision with full context
•Natural language explanations: "Why did I choose X over Y?"
•Traceable decision paths for compliance
•Timeline reconstruction of actions

4. Self-Diagnostics

•"My bid predictions have been off by 15% this week"
•"Users are rejecting 80% of my keyword recommendations"
•Proactive problem identification before user notices
•Performance degradation alerts

5. Feature Attribution

•Which features provide the most value?
•What insights are users actually using?
•Dead code and unused feature detection

Decision Audit Example

json

{
  "decision_id": "dec_20260205_001",
  "decision_type": "bid_optimization",
  "timestamp": "2026-02-05T11:00:00Z",
  "options_considered": [
    {"action": "increase_bid", "value": 1.80, "confidence": 0.72},
    {"action": "decrease_bid", "value": 1.20, "confidence": 0.65},
    {"action": "no_change", "value": 1.50, "confidence": 0.58}
  ],
  "chosen_option": "increase_bid",
  "reasoning": "Ensemble model predicted 0.72 confidence for bid increase. Historical data shows this keyword responds well to bid increases (+12% sales on average). Current ACoS of 18% is below target of 25%, allowing room for investment.",
  "confidence": 0.72,
  "actual_outcome": {
    "acos": 20.5,
    "sales_change": "+9%",
    "prediction_accuracy": 0.89
  }
}

Performance Tracking

json

{
  "model_name": "bid_optimizer_ensemble",
  "period": "last_7_days",
  "metrics": {
    "predictions_made": 1247,
    "average_confidence": 0.76,
    "actual_accuracy": 0.71,
    "calibration_error": 0.05,
    "trend": "stable"
  },
  "diagnosis": "Model is well-calibrated. Slight overconfidence (+5%) but within acceptable range.",
  "recommendation": "No action needed"
}

Self-Diagnostic Alert

json

{
  "alert_id": "diag_202602051100",
  "severity": "warning",
  "component": "keyword_recommender",
  "issue": "User rejection rate increased to 78% (baseline: 20%)",
  "context": {
    "recent_changes": ["Updated to more aggressive strategy 3 days ago"],
    "affected_campaigns": 5,
    "user_feedback_pattern": "Consistently rejecting bid increases >20%"
  },
  "hypothesis": "Recent strategy update misaligned with user's conservative preferences",
  "suggested_action": "Revert to previous strategy or consult memory-palace for user preferences"
}

API Operations

Get Decision Audit

json

{
  "action": "get_decision_audit",
  "decision_id": "dec_20260205_001"
}

Track Model Performance

json

{
  "action": "track_performance",
  "model_name": "bid_optimizer_ensemble",
  "prediction": {...},
  "actual_outcome": {...}
}

Run Self-Diagnostic

json

{
  "action": "run_diagnostic",
  "components": ["all"] // or specific: ["bid_optimizer", "anomaly_detector"]
}

Response:

json

{
  "overall_health": "good",
  "issues_found": 1,
  "diagnostics": [
    {
      "component": "keyword_recommender",
      "status": "degraded",
      "details": {...}
    }
  ]
}

Usage Patterns

Pattern 1: Post-Decision Learning

code

AFTER DECISION:
1. Log decision with full context
2. Wait for actual outcome (24-48 hours)
3. Compare prediction vs. reality
4. Update confidence calibration
5. Store in memory-palace if significant pattern

Pattern 2: Proactive Health Check

code

DAILY CRON:
1. Run self-diagnostic on all components
2. Check model accuracy trends
3. Identify degradation early
4. Alert user if critical
5. Auto-adjust if within parameters

Pattern 3: "Why Did You Do That?"

code

USER: "Why did you decrease this bid?"
CONSCIOUSNESS ENGINE:
1. Retrieve decision audit for that bid change
2. Extract reasoning and context
3. Present in natural language
4. Include confidence score and alternatives considered

Database Schema

sql

-- From server/updates/04_meta_skills_tables.sql

decision_audit (
  decision_type,
  options_considered,
  chosen_option,
  reasoning,
  confidence,
  actual_outcome,
  created_at
)

model_performance_tracking (
  model_name,
  prediction_accuracy,
  usage_count,
  avg_confidence,
  last_calibrated,
  performance_trend
)

Integration with Other Skills

Feeds data to:

•memory-palace: Store successful decision patterns
•evolution-engine: Identify underperforming strategies
•meta-learner: Adjust learning rates based on accuracy
•narrative-architect: Generate explanations for users

Files

code

.agent/skills/consciousness-engine/
├── SKILL.md
├── scripts/
│   ├── decision_logger.py           # Audit trail generation
│   ├── confidence_calibrator.py     # Accuracy tracking
│   └── performance_monitor.py       # Self-diagnostics
└── tests/
    └── test_decision_logger.py

Example Invocation

code

USER: "Why have my conversions been dropping?"

CONSCIOUSNESS ENGINE ACTION:
1. Run diagnostic on conversion-related components
2. Check recent decisions affecting conversion rate
3. Compare prediction accuracy vs. baseline
4. FIND: Bid optimization model accuracy dropped from 0.82 to 0.68
5. FIND: Recent market condition changes not captured in features
6. Report: "I detected that my bid predictions have been less accurate recently (68% vs. baseline 82%). This coincides with increased market competition that my current features don't fully capture. Recommend: Update feature engineering to include competitive density metrics."

Key Principle

The system that knows what it doesn't know is more trustworthy than one that doesn't.

Consciousness Engine ensures Optimus Pryme:

•Never silently fails
•Learns from mistakes
•Admits uncertainty
•Self-corrects proactively

This skill transforms Optimus Pryme from a "black box" into a transparent, self-aware system you can trust.