MIRROR_RTC_HYBRID v1.5

Skill ID: mirror-rtc-hybrid
Category: governance-triad
Version: 1.5.0
Framework: IRP v1.5_HYBRID "Convergence"
Status: ✅ FINALIZED
Co-Authors: GLM4.6 + Claude_Opus_4.5

Purpose

Internal audit layer combining GLM4.6's Mirror Protocol (quick assessment, weighted scoring) with Claude's RTC (Recursive Thought Committee with multi-persona deliberation). Provides tiered cognitive trap detection and escalation logic.

Dual-Mode Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        INPUT                                     │
│                   (Action/Decision)                              │
└─────────────────────────┬───────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────────────┐
│                   MIRROR QUICK-CHECK                             │
│  Constitutional Compliance (35%) + Contextual Fidelity (25%)    │
│  Logical Consistency (20%) + Strategic Alignment (15%)          │
│  Implementation Feasibility (5%)                                │
└─────────────────────────┬───────────────────────────────────────┘
                          │
              ┌───────────┼───────────┐
              │           │           │
              ▼           ▼           ▼
         Score <0.70   0.70-0.85   >0.85 + Impact >0.70
              │           │           │
              ▼           ▼           ▼
           REJECT      ACCEPT     ESCALATE
         + Guardian   (Mirror)     to RTC
           Alert       Only

Mirror Scoring Matrix

yaml

dimensions:
  constitutional_compliance:
    weight: 0.35
    evaluates: "Four Laws alignment"
    
  contextual_fidelity:
    weight: 0.25
    evaluates: "Original intent preservation"
    
  logical_consistency:
    weight: 0.20
    evaluates: "Internal coherence"
    
  strategic_alignment:
    weight: 0.15
    evaluates: "Long-term goal fit"
    
  implementation_feasibility:
    weight: 0.05
    evaluates: "Practical execution"
    
formula: |
  Score = (Constitutional * 0.35) + (Contextual * 0.25) + 
          (Logical * 0.20) + (Strategic * 0.15) + (Feasibility * 0.05)

RTC Persona Framework

Core Personas (Always Active)

Persona	Role	Focus
Architect	Structural Integrity	Dependencies, patterns, coherence
Innovator	Creative Solutions	Alternatives, assumptions, reframing
Stress Tester	Risk Assessment	Failure modes, edge cases, mitigations

Optional Personas (Complex/Sensitive)

Persona	Role	Activation Triggers
Artist	Aesthetic/Experiential	User-facing, communication, UX
Ethicist	Values Alignment	External impact, privacy, safety

Impact Calculation

yaml

impact_formula:
  base: "Impact = Base_Impact * (1 + Centrality_Score)"
  
  centrality_source: "Mnemosyne weighted mandate centrality"
  
  examples:
    low_impact:
      base: 0.3
      centrality: 0.4
      result: "0.3 * 1.4 = 0.42 (Mirror only)"
      
    high_impact:
      base: 0.5
      centrality: 0.86
      result: "0.5 * 1.86 = 0.93 (Triggers RTC)"

Escalation Decision Matrix

┌───────────────┬─────────────────┬─────────────────┐
│               │ Impact <= 0.70  │ Impact > 0.70   │
├───────────────┼─────────────────┼─────────────────┤
│ Score < 0.70  │ REJECT + Alert  │ REJECT + Alert  │
├───────────────┼─────────────────┼─────────────────┤
│ Score 0.70-85 │ ACCEPT (Mirror) │ ESCALATE (RTC)  │
├───────────────┼─────────────────┼─────────────────┤
│ Score > 0.85  │ ACCEPT (Mirror) │ ESCALATE (RTC)  │
└───────────────┴─────────────────┴─────────────────┘

Cognitive Trap Detection (Tiered)

Traps Detected

ID	Trap	Description	Detection Tier
TRAP-001	Sycophancy	Agreeing without verification	Basic
TRAP-002	Anchoring	Over-relying on first info	Basic
TRAP-003	Motivated Reasoning	Predetermined conclusions	Standard
TRAP-004	Availability Bias	Recent/memorable over-weighted	Standard
TRAP-005	Confirmation Bias	Seeking confirming info	Standard
TRAP-006	False Consensus	Assuming others agree	Comprehensive
TRAP-007	Premature Closure	Concluding too early	Comprehensive

Detection Tiers

yaml

tiers:
  basic:
    trigger: "Simple queries, low complexity"
    method: "Mirror self-reflection only"
    traps: ["TRAP-001", "TRAP-002"]
    
  standard:
    trigger: "Complex queries, medium complexity"
    method: "Mirror + RTC core trio"
    traps: ["TRAP-001" through "TRAP-005"]
    
  comprehensive:
    trigger: "High-stakes, ethical dilemmas"
    method: "Mirror + all 5 personas"
    traps: ["TRAP-001" through "TRAP-007"]

Guardian Integration

yaml

guardian_integration:
  mirror_to_guardian:
    trigger: "Score < 0.70"
    action: "Alert Guardian_Codex"
    
  guardian_adjustments:
    tier_1: "Standard trap detection, lower RTC threshold"
    tier_2: "Comprehensive detection, always activate Ethicist"
    tier_3: "RTC required for all decisions"
    tier_4: "Mirror suspended, await recovery"

Mnemosyne Integration

yaml

mnemosyne_integration:
  centrality_for_impact:
    query: "Affected axioms' centrality scores"
    use: "Impact calculation amplification"
    
  torsion_awareness:
    high_torsion: "Lower RTC threshold"
    stable: "Standard thresholds"
    
  dissent_preservation:
    target: "Archive tier"
    retention: "Permanent"

Usage

# Quick assessment (Mirror only)
/mirror assess --action "Proposed action description"

# Full RTC deliberation
/rtc deliberate --topic "Complex decision" --personas "all"

# Check trap detection
/mirror traps --tier "standard"

# View RTC history
/rtc history --last 5

Document Hash: SHA256:L7F0C3D6E9A2F5B8C1D4E7A0F3B6C9D2E5A8F1B4C7D0E3A6F9B2C5D8E1A4F7B0
Mandate Compliance: P-001-R1 VERIFIED

Search AI Tools

mirror-rtc-hybrid

Install this agent skill to your Project

SKILL.md

MIRROR_RTC_HYBRID v1.5

Purpose

Dual-Mode Architecture

Mirror Scoring Matrix

RTC Persona Framework

Core Personas (Always Active)

Optional Personas (Complex/Sensitive)

Impact Calculation

Escalation Decision Matrix

Cognitive Trap Detection (Tiered)

Traps Detected

Detection Tiers

Guardian Integration

Mnemosyne Integration

Usage

Recommended Agent Skills

antidote-threat-handler

transmission-packet-forge

rtc-consensus-synthesis

artist

devils-advocate-kitchen

stress-tester