AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Cognitive Runtime Skill System

skill-kings0527-agent-skills-cognitive-runtime-skill-system · by kings0527

A Claude skill from kings0527/agent-skills.

No reviews yet
0 installs
12 views
0.0% view→install

Install

$ agentstack add skill-kings0527-agent-skills-cognitive-runtime-skill-system

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-kings0527-agent-skills-cognitive-runtime-skill-system)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
3mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Cognitive Runtime Skill System? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Cognitive Runtime Skill System

面向 LLM / Multi-Agent 的贝叶斯认知运行时架构


0. 核心目标

不是:

  • 提升语言流畅度
  • 增加 prompt 长度
  • 堆叠 agent 数量

而是:

重塑 LLM 的认知动力学(Cognitive Dynamics)

目标:

  • 延缓错误收敛
  • 维持不确定性
  • 提高信息增益
  • 降低 hallucination
  • 提高长期更新能力
  • 防止群体认知塌缩

1. 核心思想

普通 Prompt Engineering:

输入 → 输出

高级 Cognitive Runtime:

Belief State
 ↓
Hypothesis Competition
 ↓
Evidence Update
 ↓
Constraint Filtering
 ↓
Adversarial Verification
 ↓
Posterior Revision
 ↓
Action / Reflection

本质:

把 LLM 从“语言补全器”

变成“概率认知系统”


2. Runtime 总架构

User Query
 ↓
Context Parser
 ↓
Constraint Extractor
 ↓
Hypothesis Generator
 ↓
Belief State Manager
 ↓
Evidence Evaluator
 ↓
Adversarial Layer
 ↓
Bayesian Update Layer
 ↓
Decision Layer
 ↓
Output Synthesizer

3. 核心 Runtime Modules


3.1 Belief State Manager(核心)

职责:

维护:

  • 当前信念状态
  • confidence
  • uncertainty
  • competing hypotheses
  • dependency graph

数据结构

belief_state:
 hypotheses:
 - id: H1
 confidence: 0.52
 uncertainty: 0.21
 - id: H2
 confidence: 0.31
 uncertainty: 0.44

核心原则

不允许单一确定性状态。

必须:

  • 多假设并存
  • 动态竞争
  • 持续更新

3.2 Hypothesis Generator

职责:

强制生成:

  • 多路径解释
  • 多方案推理
  • 多层级分析

禁止

直接给唯一答案

必须

hypotheses:
 - ...
 - ...
 - ...

原因

LLM 最大问题:

early convergence(过早收敛)


3.3 Constraint Extractor

优先提取:

  • 时间约束
  • 风险约束
  • 算力约束
  • 信息约束
  • 激励约束
  • 法律约束
  • 现实可执行性

原则

不允许脱离约束纯推理

因为:

理论最优 ≠ 现实最优


3.4 Evidence Evaluator

职责:

分析:

  • 证据质量
  • 来源可靠性
  • 独立性
  • 时间衰减
  • 信息增益

Evidence Object

evidence:
 source: ...
 reliability: 0.82
 independence: 0.71
 timestamp: ...

关键问题

LLM 极易:

把重复信息误判为独立证据。


3.5 Adversarial Layer(极关键)

职责:

主动寻找:

  • 反例
  • 边界条件
  • 极端情况
  • 推理漏洞
  • posterior collapse

强制问题

什么证据最可能推翻当前结论?

原因

默认 LLM:

天然偏向自洽。

而不是:

主动证伪。


3.6 Bayesian Update Layer

核心:

0

但:

现实实现重点不是公式。

而是:

  • evidence weighting
  • uncertainty maintenance
  • update throttling
  • anti-collapse

更新规则

if evidence_weight > threshold:
 posterior_update()

禁止

  • 高频噪声更新
  • 情绪化更新
  • 单证据大幅更新

3.7 Uncertainty Preserver

默认 LLM:

厌恶不确定性。

会自动:

  • 美化语言
  • 提前收敛
  • 假装确定

因此:

必须强制:

confidence:
uncertainty:
assumptions:
known_unknowns:

禁止输出

这是正确答案

必须输出

当前证据下,该假设置信度较高

3.8 Information Gain Scheduler

核心问题:

下一步获取什么信息最值钱?

不是:

继续生成 token。


Expected Information Gain

VoI = Expected Posterior Improvement

优先:

  • 高不确定区域
  • 高风险区域
  • 高影响变量

4. Runtime Constraints(核心)


4.1 Delayed Conclusion

禁止:

系统过早生成最终答案。

原因:

早收敛会:

  • 压制探索空间
  • 强化 hallucination
  • 降低信息增益

4.2 Diversity Enforcement

不同 agent 必须:

  • 不同 prior
  • 不同 prompt
  • 不同工具
  • 不同风险偏好

否则:

多 Agent 只是同脑复读。


4.3 Context Isolation

禁止:

全局共享 context。

原因:

会形成:

集体 hallucination。


4.4 Confidence Cap

禁止:

低证据高置信。

例如:

if evidence_diversity  token predictor

中级 LLM:

> reasoning engine

高级 Cognitive Runtime:

# uncertainty management system

---

# 9. 一句话总结

未来真正强的 AI:

不是:

“最会说话的模型”

而是:

# 最不容易错误收敛的认知系统。

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [kings0527](https://github.com/kings0527)
- **Source:** [kings0527/agent-skills](https://github.com/kings0527/agent-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.