README.md

metadata

title: Ghost Malone
emoji: 👻
colorFrom: purple
colorTo: blue
sdk: docker
pinned: false

Ghost Malone: Emotion-Aware MCP System

📢 Project Announcement: https://x.com/badtrapazoid/status/1990539436795572375 🌐 Live Space: https://huggingface.co/spaces/MCP-1st-Birthday/ghostMalone

A three-server emotional engine built for the MCP ecosystem.

What It Does

Ghost Malone turns raw text into a structured emotional understanding, remembers past patterns, and produces responses that fit the user’s state. Not a chatbot — a small, disciplined mind.

Architecture (Three Servers) USER → Orchestrator → Emotion Server → Memory Server → Reflection Server → Output

1. Emotion Server

Russell’s Circumplex (valence + arousal)

Fast pattern matching across 8 affect states

Outputs: labels, valence, arousal, tone

~31ms latency

2. Memory Server

Rolling 50-entry history

Stores text + emotional metadata

Recalls past patterns for personalization

~66ms latency

3. Reflection Server

Claude-driven tone adaptation

Uses emotion + memory + need to shape response

~5.3s latency (dominant cost)

Two Lexicons (Core Intelligence) Needs Lexicon

5 core needs: autonomy, connection, security, rest, recognition

24 context patterns → 47 inference rules

Aligns emotion with human motive

95.2% accuracy vs BPNSFS scale

Intervention Lexicon

Evidence-based strategies

Constitutional gating:

confidence ≥ 0.70

arousal ≥ 0.40

depth ≥ 2 messages

Prevents overstepping / unsolicited advice

Pipeline (Six Steps)

Emotion analysis

Needs inference

Memory recall + store

Reflection (tone-aware response)

Intervention check

Response assembly

Total latency: ~5.5s.

What Makes It Different

1. Needs, not just emotions

“Sad” branches to different needs (connection vs autonomy vs security).

2. Memory-aware

Responses reference earlier feelings.

3. Constitutional alignment

No forced advice. No toxic positivity. User controls sensitivity via sliders.

4. Tunable thresholds

Real-time control of intervention behavior.

5. Emotional trajectory visualization

Simple plot showing how the user is moving on the Circumplex.

Core Example (One Glance)

Input: “I feel so isolated and alone.”

Emotion: sad, lonely (valence -0.6, arousal 0.4)

Need: connection (0.92)

Memory: user mentioned “feeling left out at work”

Response: grounded, gentle reflection

Intervention (if gated): connection strategies

README Example Section

## 🎯 Try These Examples

**For Connection needs:**
1. "I feel so isolated and alone"
2. "Nobody really understands what I'm going through"

**For Autonomy needs:**
1. "I feel so trapped and powerless in this situation"
2. "I have no control over anything anymore"

**For Security needs:**
1. "Everything feels so uncertain and scary"
2. "I'm worried about what's going to happen next"

**For Rest needs:**
1. "I'm so burnt out and exhausted"
2. "I'm completely drained and can't keep going"

**For Recognition needs:**
1. "Nobody notices all the work I do"
2. "I feel completely invisible and unappreciated"

💡 Interventions appear on the 2nd message when thresholds are met.