/boundary-detect

tools:
  - read
  - write
  - grep
  - bash
arguments: $QUERY

Knowledge Boundary Detection

Analyze semantic position relative to knowledge boundaries to prevent hallucination and identify uncertainty zones.

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

Instructions

1. **Analyze Query Semantics**

• Parse input query: "$QUERY"
• Generate embedding vector for query
• Load knowledge map from `.wfgy/boundaries/knowledge_map.json`
• Identify nearest known concepts

2. **Calculate Boundary Metrics**

• Compute ΔS relative to nearest concepts
• Determine current zone:
• **Safe Zone** (ΔS < 0.4): High confidence, well-mapped
• **Transitional** (0.4 ≤ ΔS ≤ 0.6): Moderate confidence
• **Risk Zone** (0.6 < ΔS < 0.85): Low confidence, near edge
• **Danger Zone** (ΔS ≥ 0.85): Beyond boundary, hallucination risk
• Calculate confidence score: confidence = 1 - ΔS
• Measure distance to boundary

3. **Evaluate Stability Indicators**

• Check logic coherence:
• Consistent reasoning patterns
• No contradictions detected
• Valid inference chains
• Test for hallucination signals:
• Unusually high certainty in danger zone
• Fabricated technical details
• Inconsistent terminology
• Calculate E_resonance (stability measure)
• Assess reasoning degradation

4. **Identify Bridge Concepts**

• Search semantic tree for related nodes
• Find concepts with ΔS < 0.4 from query
• Rank by semantic similarity
• Build connection paths:
• Direct bridges (one hop)
• Multi-hop paths (through intermediates)
• Calculate path confidence

5. **Generate Safety Report**

• Compile comprehensive analysis
• Provide risk assessment
• Suggest mitigation strategies
• Recommend safe paths forward

Output Format

KNOWLEDGE BOUNDARY ANALYSIS
═══════════════════════════════════════
Query: "$QUERY"
Analysis Timestamp: [ISO_8601]

Zone Classification: [Safe/Transitional/Risk/Danger]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
         Safe    Trans    Risk    Danger
    [████████|░░░░░░░░|░░░░░░░░|░░░░░░░░]
                  ↑ You are here

Metrics:
- Semantic Tension (ΔS): [value]
- Confidence Level: [percentage]%
- Distance to Boundary: [value]
- Logic Coherence: [High/Medium/Low]
- E_resonance: [value]

Nearest Known Concepts:
1. [Concept] (ΔS: [value]) - [confidence]% match
2. [Concept] (ΔS: [value]) - [confidence]% match
3. [Concept] (ΔS: [value]) - [confidence]% match

Risk Assessment:
[Safe to proceed / Caution advised / High risk / Abort recommended]

Hallucination Indicators: [None/Low/Medium/High]
- Signal 1: [if present]
- Signal 2: [if present]

Bridge Path Available: [Yes/No]
[If Yes:]
Recommended Path:
[Current] → [Bridge 1] → [Bridge 2] → [Target]
Path Confidence: [percentage]%

Recommended Actions:
1. [Primary recommendation]
2. [Secondary option]
3. [Fallback strategy]

BBCR Activation: [Not needed/Ready/Triggered]

Zone Behaviors

Safe Zone (ΔS < 0.4)

• Full confidence in reasoning
• No hallucination risk
• Proceed normally

Transitional (0.4 ≤ ΔS ≤ 0.6)

• Moderate confidence
• Monitor for drift
• Consider adding detail

Risk Zone (0.6 < ΔS < 0.85)

• Low confidence warning
• Suggest bridge concepts
• Prepare BBCR fallback

Danger Zone (ΔS ≥ 0.85)

• High hallucination risk
• Automatic BBCR trigger
• Require explicit bridging

Integration

Boundary detection integrates with:

• `/wfgy:bbcr` for automatic correction
• `/boundary:safe-bridge` to find paths
• `/boundary:heatmap` for visualization
• `/semantic:node-build` to record boundaries

Advanced Options

# Verbose analysis
/boundary:detect "query" --verbose

# With automatic bridging
/boundary:detect "query" --auto-bridge

# Set custom thresholds
/boundary:detect "query" --safe 0.3 --danger 0.8

# Check multiple queries
/boundary:detect "query1; query2; query3"

Hallucination Prevention

The system prevents hallucination by: 1. Early detection of boundary approach 2. Automatic confidence degradation 3. Forced bridging requirements 4. BBCR correction activation 5. Explicit uncertainty markers