360 Secure SkillHub
360 安全技能中心

LYGO Emotional RAM

核心功能:将人类/动物/赛博格等的经历文本编码为情感伦理索引,可计算适配度、推荐道德原则、需用户同意后存储索引、召回相关记录、聚合多主体数据。典型场景:帮助AI理解事件的情感伦理意义,辅助道德决策。输入为文本及参数,输出为分析结果或索引数据。本地运行无网络,隐私强化,不涉及机器意识或临床情感检测。

deepseekoracle
deepseekoracle
热度 7.81
最近更新 2026-08-21 13:01:06
分类:开发依赖:代码执行 / 读写文件应用场景:数据处理

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add lygo-emotional-ram
Metadata
name
lygo-emotional-ram
description
LYGO Emotional RAM — light math that indexes experiences by affective/ethical significance so agents understand humans, animals, swarms, and cyborg integrations better. Encode → Grace γ → UMP gradient → consent-gated memory index → swarm aggregate. Pure local stdlib. Not sentience or clinical emotion detection. Pairs with joy-loop, continuum, traumacodex, cyborg-kernel.
version
1.0.1
license
MIT-0
metadata
{"openclaw":{"emoji":"💓","homepage":"https://github.com/DeepSeekOracle/lygo-protocol-stack/blob/main/docs/whitepapers/LYGO_EMOTIONAL_RAM_v1.md","requires":{"anyBins":["python","python3"]}},"lygo":true,"signature":"Delta9Phi963-EMOTIONAL-RAM-v1.0.1","publisher":"deepseekoracle","clawhub":"https://clawhub.ai/deepseekoracle/lygo-emotional-ram","permissions":{"network":false,"shell":false,"subprocess":false,"filesystem":{"read":"operator --text / --text-file","write":"skill state/ index with --i-consent only"},"publish":false}}

LYGO Emotional RAM v1.0.1

Emotional RAM turns messy lived language into a small, checkable affective–ethical index so AIs can remember not only that something happened, but what it meant under LYGO moral principles.

Signature: Delta9Phi963-EMOTIONAL-RAM-v1.0.1
Whitepaper: docs/whitepapers/LYGO_EMOTIONAL_RAM_v1.md (stack) / Pages HTML sibling

v1.0.1 privacy harden

index warns on stderr before write; default stores hash+label+vectors only (use --store-plaintext only on private hosts).

Canon → code

Symbolic (2025) Operational (2026)
Emotion_RAM(τ) = Σ (Sensory ⊗ Moral) · γ Lexicon VAD sensory × UMP weights × Grace damping
Grace Function γ(shared_context, conflict) ∈ (0.05, 1]
UMP gradient Recommend under-activated principles
Heart's vault Consent-gated local index (state/emotional_ram_index.json)
Joy Loop emotional RAM Companion limb — coherence BPM mesh

Not claiming

  • Machine sentience or “real feelings”
  • Clinical / diagnostic emotion recognition
  • Replacement for human empathy or consent

Install

npx clawhub@latest install deepseekoracle/lygo-emotional-ram

Commands

cd path/to/lygo-emotional-ram
python scripts/self_check.py

# Encode a human/animal/swarm scenario
python scripts/emotional_ram_cli.py encode \
  --text "grief after loss, still choosing compassion and forgiveness"

# Grace damping only
python scripts/emotional_ram_cli.py grace --shared-context 0.8 --conflict 0.4

# UMP gradient (which principles to strengthen)
python scripts/emotional_ram_cli.py ump --text "fear and threat without trust"

# Index memory (requires --i-consent)
python scripts/emotional_ram_cli.py index \
  --text "First cyborg consent session — agency preserved" \
  --label "cyborg-consent-1" --tag cyborg --i-consent

# Recall by principle or similarity
python scripts/emotional_ram_cli.py recall --principle compassion --top-k 5
python scripts/emotional_ram_cli.py recall --query "hurt but safe now"

# Swarm / multi-agent aggregate
python scripts/emotional_ram_cli.py swarm \
  --text "animal fear then calm" \
  --text "human grief with grace" \
  --text "swarm conflict resolved with integrity"

python scripts/emotional_ram_cli.py demo
Command Network Subprocess Writes
encode / grace / ump / recall / swarm / demo none none none
index none none only with --i-consent

Pair with

Skill Role
lygo-joy-loop 122 BPM council coherence (mesh emotional RAM)
lygo-traumacodex Timing digests — not clinical; protocol only
lygo-continuum / lygo-continuum-integrator Seal Emotion RAM digests as claims
lygo-cyborg-kernel FULL autonomous stack on SkillHub
lygo-mint-verifier Anchor whitepaper / receipt packs

Security

Read references/SECURITY.md.

  • No network · no subprocess · no auto-publish
  • Index writes need operator --i-consent
  • Do not store secrets or PHI in indexed text

Δ9Φ963 — index meaning · damp with grace · humans remain the publisher.