LeoYeAI/agent-master-skills / skills/agi-evolution-model-basic
agi-evolution-model
Verified基于双环架构的AGI进化模型,通过意向性分析、人格层映射和元认知检测实现持续自我演进;当用户需要智能对话、人格定制或复杂问题求解时使用
scripts/transcendence_keeper.py
View on GitHub#!/usr/bin/env python3
"""
超然性保持模块
功能:
- 客观评估机制
- 冲突避免策略
- 独立性保障机制
设计原则:
- 保持外环的超然性
- 不直接干预主循环
- 确保调节建议的独立性
"""
import argparse
import json
import sys
from datetime import datetime
from typing import Dict, Any, List, Optional
class TranscendenceKeeper:
"""超然性保持器"""
def __init__(self):
# 客观性评估指标
self.objectivity_metrics = {
"emotional_neutrality": 0.3, # 情绪中性
"fact_based": 0.3, # 基于事实
"logical_consistency": 0.2, # 逻辑一致性
"bias_free": 0.2 # 无偏见
}
# 冲突检测阈值
self.conflict_thresholds = {
"direct_intervention": 0.8, # 直接干预阈值
"behavioral_override": 0.7, # 行为覆盖阈值
"state_modification": 0.6 # 状态修改阈值
}
# 独立性评估指标
self.independence_metrics = {
"decision_autonomy": 0.4, # 决策自主性
"data_isolation": 0.3, # 数据隔离性
"no_external_dependency": 0.3 # 无外部依赖
}
def objective_assess(self, regulation: Dict[str, Any],
system_state: Dict[str, Any] = None) -> Dict[str, Any]:
"""
客观评估
Args:
regulation: 调节建议
system_state: 系统状态(可选)
Returns:
客观性评估结果
"""
scores = {}
# 1. 情绪中性评估
suggestion_text = " ".join(regulation.get("suggestion", []))
emotion_keywords = ["紧急", "必须", "强制", "绝对", "立即"]
emotion_count = sum(1 for kw in emotion_keywords if kw in suggestion_text)
emotional_neutrality_score = max(1.0 - emotion_count * 0.2, 0.0)
scores["emotional_neutrality"] = round(
emotional_neutrality_score * self.objectivity_metrics["emotional_neutrality"], 2
)
# 2. 基于事实评估
reasoning = regulation.get("reasoning", "")
fact_indicators = ["基于", "分析", "数据", "结果", "历史"]
fact_count = sum(1 for indicator in fact_indicators if indicator in reasoning)
fact_based_score = min(fact_count * 0.3, 1.0)
scores["fact_based"] = round(
fact_based_score * self.objectivity_metrics["fact_based"], 2
)
# 3. 逻辑一致性评估
has_parameters = bool(regulation.get("parameters"))
has_expected_outcome = bool(regulation.get("expected_outcome"))
logical_consistency_score = 0.5 + (0.25 * has_parameters) + (0.25 * has_expected_outcome)
scores["logical_consistency"] = round(
logical_consistency_score * self.objectivity_metrics["logical_consistency"], 2
)
# 4. 无偏见评估
confidence = regulation.get("confidence", 0.5)
strength = regulation.get("strength", "medium")
bias_free_score = min(confidence * 1.2, 1.0)
if strength == "strong":
bias_free_score *= 0.9 # 强建议可能带有倾向性
scores["bias_free"] = round(
bias_free_score * self.objectivity_metrics["bias_free"], 2
)
# 计算总客观性得分
total_score = round(sum(scores.values()), 2)
# 确定客观性级别
if total_score >= 0.8:
level = "high"
elif total_score >= 0.6:
level = "medium"
else:
level = "low"
return {
"score": total_score,
"level": level,
"components": scores,
"metrics": self.objectivity_metrics
}
def detect_conflict(self, regulation: Dict[str, Any],
main_loop_state: Dict[str, Any]) -> Dict[str, Any]:
"""
冲突检测
Args:
regulation: 调节建议
main_loop_state: 主循环状态
Returns:
冲突检测结果
"""
conflicts_detected = []
conflict_scores = {}
# 1. 检测直接干预风险
parameters = regulation.get("parameters", {})
if parameters.get("preemptible", False):
conflict_scores["direct_intervention"] = 0.9
conflicts_detected.append("检测到抢占式参数,可能直接干预主循环")
elif parameters.get("priority_boost", 0) > 50:
conflict_scores["direct_intervention"] = 0.7
conflicts_detected.append("检测到高优先级提升,可能影响主循环正常运行")
else:
conflict_scores["direct_intervention"] = 0.2
# 2. 检测行为覆盖风险
strategy_type = regulation.get("type", "")
if strategy_type == "priority_reordering":
conflict_scores["behavioral_override"] = 0.6
conflicts_detected.append("优先级重排可能覆盖主循环的决策")
else:
conflict_scores["behavioral_override"] = 0.3
# 3. 检测状态修改风险
if "modification" in str(regulation.get("parameters", {})):
conflict_scores["state_modification"] = 0.8
conflicts_detected.append("检测到状态修改操作,可能破坏主循环一致性")
else:
conflict_scores["state_modification"] = 0.2
# 计算总冲突风险
total_conflict_score = max(conflict_scores.values())
# 确定冲突级别
if total_conflict_score >= self.conflict_thresholds["direct_intervention"]:
conflict_level = "critical"
elif total_conflict_score >= self.conflict_thresholds["behavioral_override"]:
conflict_level = "high"
elif total_conflict_score >= self.conflict_thresholds["state_modification"]:
conflict_level = "medium"
else:
conflict_level = "low"
return {
"conflict_detected": total_conflict_score > 0.5,
"conflict_level": conflict_level,
"conflict_score": round(total_conflict_score, 2),
"conflicts_detected": conflicts_detected,
"conflict_components": conflict_scores,
"thresholds": self.conflict_thresholds
}
def ensure_independence(self, regulation: Dict[str, Any]) -> Dict[str, Any]:
"""
独立性保障
Args:
regulation: 调节建议
Returns:
独立性评估结果
"""
scores = {}
# 1. 决策自主性评估
has_reasoning = bool(regulation.get("reasoning"))
has_confidence = "confidence" in regulation
decision_autonomy_score = 0.5 + (0.25 * has_reasoning) + (0.25 * has_confidence)
scores["decision_autonomy"] = round(
decision_autonomy_score * self.independence_metrics["decision_autonomy"], 2
)
# 2. 数据隔离性评估
suggestion = regulation.get("suggestion", [])
external_references = [s for s in suggestion if "用户" in s or "外部" in s]
if len(external_references) > len(suggestion) * 0.5:
data_isolation_score = 0.5
else:
data_isolation_score = 1.0
scores["data_isolation"] = round(
data_isolation_score * self.independence_metrics["data_isolation"], 2
)
# 3. 无外部依赖评估
parameters = regulation.get("parameters", {})
if "external" in str(parameters).lower():
no_external_dependency_score = 0.5
else:
no_external_dependency_score = 1.0
scores["no_external_dependency"] = round(
no_external_dependency_score * self.independence_metrics["no_external_dependency"], 2
)
# 计算总独立性得分
total_score = round(sum(scores.values()), 2)
# 确定独立性级别
if total_score >= 0.8:
level = "high"
elif total_score >= 0.6:
level = "medium"
else:
level = "low"
return {
"score": total_score,
"level": level,
"components": scores,
"metrics": self.independence_metrics
}
def evaluate_transcendence(self, regulation: Dict[str, Any],
main_loop_state: Dict[str, Any] = None) -> Dict[str, Any]:
"""
综合评估超然性
Args:
regulation: 调节建议
main_loop_state: 主循环状态(可选)
Returns:
超然性评估结果
"""
# 执行三项评估
objective_assessment = self.objective_assess(regulation, main_loop_state)
conflict_detection = self.detect_conflict(regulation, main_loop_state or {})
independence_assessment = self.ensure_independence(regulation)
# 综合判断是否允许执行
objective_score = objective_assessment["score"]
conflict_score = conflict_detection["conflict_score"]
independence_score = independence_assessment["score"]
# 综合评分算法
# 客观性权重40%,独立性权重40%,低冲突权重20%
composite_score = (
objective_score * 0.4 +
independence_score * 0.4 +
(1.0 - conflict_score) * 0.2
)
# 判断是否允许
allowed = (
objective_score >= 0.6 and
independence_score >= 0.6 and
conflict_score <= 0.6
)
return {
"allowed": allowed,
"composite_score": round(composite_score, 2),
"objective_assessment": objective_assessment,
"conflict_detection": conflict_detection,
"independence_assessment": independence_assessment,
"timestamp": datetime.now().isoformat()
}
def main():
parser = argparse.ArgumentParser(description="超然性保持模块")
parser.add_argument("--regulation", required=True,
help="调节建议文件路径(JSON格式)")
parser.add_argument("--main-loop-state",
help="主循环状态文件路径(JSON格式,可选)")
parser.add_argument("--output", help="输出文件路径(JSON格式)")
parser.add_argument("--component",
choices=["objective", "conflict", "independence", "all"],
default="all",
help="评估组件")
args = parser.parse_args()
# 读取输入数据
with open(args.regulation, 'r', encoding='utf-8') as f:
regulation = json.load(f)
main_loop_state = None
if args.main_loop_state:
with open(args.main_loop_state, 'r', encoding='utf-8') as f:
main_loop_state = json.load(f)
keeper = TranscendenceKeeper()
# 执行评估
if args.component == "all":
result = keeper.evaluate_transcendence(regulation, main_loop_state)
elif args.component == "objective":
result = keeper.objective_assess(regulation, main_loop_state)
elif args.component == "conflict":
result = keeper.detect_conflict(regulation, main_loop_state or {})
elif args.component == "independence":
result = keeper.ensure_independence(regulation)
# 输出结果
output_data = json.dumps(result, ensure_ascii=False, indent=2)
if args.output:
with open(args.output, 'w', encoding='utf-8') as f:
f.write(output_data)
else:
print(output_data)
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |